DELFT UNIVERSITY OF TECHNOLOGY

Automating AC Power Flow Simulations

The European Electricity Grid

Interim Thesis Report

Supervisors TU Delft:
Prof.dr.ir. C. Vuik
Author: M.E. Kootte
Shravan Chipli Supervisors TenneT:
Martin Wevers
Jorrit Bos

February 2021
 Foreword

This MSc graduation project is being done in the group of Numerical Mathematics, TU
Delft, in collaboration with TenneT TSO B.V. TenneT is a European electricity trans-
mission system operator (TSO) that manages the high-voltage grid in the Netherlands
and large parts of Germany.

The power flow problem or the load flow problem involves determining the voltage
magnitudes and phase angles in every bus of the power system. These quantities are
then used to compute the power flow in every branch of the power system network.
This constitutes the steady state power flow simulations which, for given generation
and consumption, give insight into the steady state behavior of the network. Hence,
power flow simulations play a fundamental role in various sectors of a TSO such as
operation and planning.

In this interim thesis report, we present the findings of a literature study conducted in
order to understand the theory behind solving the power flow problem and to hence
achieve the research objectives of this project.

i
 Contents

1 Introduction 1

2 The Electric Grid 4

2.1 Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.2 Transmission and Distribution . . . . . . . . . . . . . . . . . . . . . . 5
2.3 Consumption . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.4 Challenges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

3 Modelling the Transmission System 7

3.1 Fundamentals of AC circuits . . . . . . . . . . . . . . . . . . . . . . . 7
3.2 Network Topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
3.3 The Power Flow Model . . . . . . . . . . . . . . . . . . . . . . . . . . 14

4 Power Flow Solvers 17

4.1 Newton-Raphson . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
4.2 Fast Decoupled Load Flow . . . . . . . . . . . . . . . . . . . . . . . . 25
4.3 DC approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
4.4 Gauss-Seidel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
4.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

5 Software Packages 36
5.1 PowerFactory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
5.2 PSS® E . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
5.3 pandapower . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

6 Conclusion 39

Bibliography 43

ii
 Chapter 1

Introduction

The goal of this project is to develop an AC power flow solver for the European elec-
tricity transmission network that includes the Netherlands, Germany, Belgium, Luxem-
bourg and France (fig. 1.1 shows TenneT’s high voltage grid), and automate calculations
in order to speed up tasks such as voltage regulation and reactive power compensation
assessments for transmission system planning. In the context of year-round calculations,
a comparison between AC and DC approaches of solving the power flow problem will
be made.

Motivation
A TSO is responsible for safe and reliable transport of electric power. One of the key
challenges to ensure safety and reliability is to keep the voltages within safe limits across
the transmission network. The rapid increase in addition of Renewable Energy Sources
(RES) into electricity networks causes voltage fluctuations to occur more frequently.
For instance, excess wind and solar generation would cause the voltages in the trans-
mission network to increase beyond the safe operating limit. To manage such situations
best in the future, clever planning of the essential infrastructure is necessary.

It is best practice for a TSO to keep the voltages within limits just by using reactive
power compensating assets such as shunt reactors and capacitor banks in normal op-
erating conditions. However, extreme operating conditions call for additional reactive
power support from generators, which is expensive and generally not desired. It seems
quite rational to anticipate abnormal operating conditions considering the stochastic na-
ture of growing renewable energy sources and extreme weather conditions. Hence, it is
very important for a TSO to efficiently plan for reactive power compensating assets for
the decades to come.

AC year-round simulations help to estimate the need for reactive power compensating
assets. TenneT currently uses commercial software packages for power flow analysis.
With increasing problem sizes and widely varying inputs, it has been found that the grid

1
models often run into convergence issues when solving the AC power flow problem.
This means that the power flow problem could be ill-posed. Ensuring convergence and
voltage regulation currently involve manual intervention. That is, manually adding ex-
ternal grids to achieve reactive power balance in the network or to change the voltage
setpoints of generators. This is a trial and error technique that could fix the ill-posedness
of the problem and get the model to converge. Once the model converges, voltages are
kept within limits by manually switching shunt reactors or capacitor banks on/off. This
is largely time consuming, especially for year-round simulations, wherein there could
possibly be convergence issues in almost all hours of the year. Hence AC power flow
simulations are currently done only for a few worst-case hours. In order to do AC year-
round simulations, an automated solver that can fix convergence issues and regulate
voltages without manual intervention is necessary. Moreover, the exact reason for the
solvers to diverge is not yet known and most often, the tricks employed to get a grid
model to converge do not work for a different model or for a different set of inputs.

In this project, we investigate convergence problems in detail and develop mathemat-

ical models that automatically ensure convergence and regulate voltages. The models
could form the basis for a potential software subroutine that can perform AC year-round
simulations for a wide range of inputs, without manual intervention.

Another important application of power flow simulations is capacity planning. As the

name implies, capacity planning involves planning the transmission system infrastruc-
ture such as additional cables in the network for the future. Year-round power flow
simulations are carried out in order to determine transmission line loading and thus plan
the additional capacities required to operate the transmission system without overload-
ing the lines in the network. TenneT currently uses the DC power flow approach to
do year-round simulations for capacity planning. The DC approach is often favored
for computational benefits and when AC power flow simulations cannot be done, for
instance, due to convergence issues. However, since the DC approach linearizes the
power flow problem, there could be a significant reduction in accuracy of the solution.
For line loading, this means that the determined values could be less than that in re-
ality, possibly leading to an inferior estimate of capacity requirements. In this project
we quantify the reduction in accuracy by comparing the solution of the automated AC
power flow simulation model with the DC solution. The advantages of using the AC
power flow model for capacity planning, if any, will be weighed by a cost analysis.

Literature review
In order to solve the aforementioned problems, we start with a literature review that
serves as a foundation to this project. The following four concepts of power flow anal-
ysis are studied.
• The power flow problem
• Power flow solvers

2
 Endrup (Denmark)
NordLink (Norway) Kassø (Denmark)
Kassø/Ensted (Denmark)
SylWin1 Kassø (Denmark)
Klixbüll_Süd
Jardelund

Husum_Nord
Schuby_West Baltic cable (Sweden)
BorWin2 BorWin3
BorWin1 HelWin2 Göhl
HelWin1
Audorf_Süd
Heide_West
DolWin6
Ten noorden van DolWin1
de Waddeneilanden DolWin2
alpha
DolWin3 ventus
Büttel
COBRAcable (Denmark)
Nordergründe
NorNed (Norway) Hamburg_Nord
Riffgat
Halbemond Stade_West
Fedderwarden Görries (Germany)
Emden_Ost Dollern
Conneforde
Eemshaven
Diele Elsfleth_West Sottrum

Loep ontwerp_2006.V16
Meeden
Ganderkesee
Dörpen_West

Hollandse Kust
IJmuiden Ver Wehrendorf (Germany)
(noord) Alpha Niederrhein Meppen
Ens Wehrendorf Landesbergen
(Germany) (Germany) (Germany)
Hollandse Kust (west) Lehrte
Zwolle Merzen (Germany)
Loep ontwerp_2006.V16

Beverwijk Wahle

Hengelo Gronau (Germany) Wolmirstedt (Germany)

Hollandse Kust (zuid) Alpha Vijfhuizen Diemen
Hollandse Kust (zuid) Beta
Arnhem
Bleiswijk
Westerlee Dodewaard Gütersloh (Germany) Grohnde
BritNed (Great-Britain) Gütersloh (Germany)
Maasvlakte Niederrhein (Germany)
Geertruidenberg

Borssele Beta
Borssele Alpha
Nehden (Germany)

Borssele
Van Eyck (Belgium) Borken
Loep ontwerp_2006.V16

Zandvliet (Belgium) Maasbracht Mecklar Vieselbach (Germany)

Loep ontwerp_2006.V16
Oberzier (Germany) Loep ontwerp_2006.V16

Siersdorf (Germany) Eisenach (Germany)

Loep ontwerp_2006.V16

Dauersberg (Germany) Altenfeld

Remptendorf (Germany)
(Germany)

Redwitz
Kriftel (Germany)
Großkrotzenburg Grafenrheinfeld

Urberach (Germany) Bayreuth

Dettingen (Germany) Aschaffenburg

Hradec (Czech Republic)

Höpfingen (Germany) Prestice (Czech Republic)

Kupferzell (Germany) Raitersaich

Schwandorf

Pleinting
Irsching
Isar
Meitingen (Germany)

Ottenhofen
APG Austria (Austria)

Sellrain Silz (Austria)

Loep ontwerp_2006.V16
Loep ontwerp_2006.V16

Figure 1.1: Grid map of TenneT

• Convergence properties of solvers

• Comparison of AC and DC approaches

The mathematical model of the transmission system is derived and studied in order to
understand the power flow problem which lies at the heart of power flow analysis. To
derive the mathematical model, we study the fundamentals of AC circuits and explore
the transmission network topology. As solving the power flow problem requires a good
understanding of the available power flow solvers, we investigate some of the promi-
nent power flow solvers used in power flow analysis and understand their capabilities.
The key to automate calculations is understanding the convergence properties of power
flow solvers. We review some of the methods available in the literature that improve
convergence of solvers. The feasibility of using the DC approximation method in place
of a full power flow solver is investigated. A few software packages commonly used
in the industry are introduced. The report starts with a brief introduction to electricity
grids and concludes with the findings of the literature review and the research questions
of the project.

3
 Chapter 2

The Electric Grid

Electric grids are some of the largest networks humanity has ever built. For more than
a hundred years1 now, they have been doing an incredible job lighting up our societies.
It’s very hard to imagine life without electricity, for the world then would come to a halt.

The fact that electric power is just a switch away in most parts of the world today is
a result of the intricate electric grids that quite seamlessly bridge generation and con-
sumption. The complexity of electric grids is increasing at a rapid pace with integration
of renewable energy sources. A strong mathematical approach and collective effort are
imperative to understand, model, and control the electric grids of this day better.

This chapter briefly describes the fundamental stages of an electric grid and a few chal-
lenges down the road.

2.1 Generation
Conventionally, electric power is generated as Alternating Current (AC) at places known
as power plants or generating stations. An AC generator transforms mechanical energy
to electric energy by electromagnetic induction. The prime movers that drive generators
could be steam/gas/water/wind turbines, internal combustion engines or nuclear reac-
tors depending on location and availability of resources. For example, electricity in the
Netherlands is produced primarily from natural gas and coal [2].

An exception to AC generation is the use of photovoltaic (PV) cells. PV cells transform

solar energy to Direct Current (DC) which is then converted to AC by an inverter before
injecting it into the grid. Solar farms are gaining ground in places with more sunlight
across the world.

1
World’s first central electric generating station - The Pearl Street Station, began operation on 4
September 1882 [1].

4
Electric power is generated at a frequency of 50 or 60 Hz [3]. This has been widely
accepted as the optimal frequency considering several applications on a wide scale over
a long period of time. However, railways, airplanes, ships and oil rigs use different
frequencies. Higher frequency systems have the advantage of a better power to weight
ratio for machines [3].

2.2 Transmission and Distribution

The distinct advantage of alternating current is that it can be efficiently stepped up and
down in voltage using a transformer [4]. High voltage power is preferred for transmis-
sion over large distances as resistive losses are less. The high voltage network called
the transmission system is responsible for transporting electricity in the order of 110-380
kV from power plants to substations [5]. The rest of the grid constituting of medium
and low voltage lines is called the distribution system and is responsible for distributing
electric power to end users. The substations shift the voltage down to the order of 10-20
kV and finally distribution transformers that are locally installed step it further down to
the order of utilization voltage (<1kV) making it suitable for domestic and commercial
use. It should also be noted that high voltage DC (HVDC) is sometimes preferred to AC
when transmission distances are long enough to justify a reduction in cost of conductors
(three phase AC needs three cables whereas DC needs two) over AC/DC conversion
costs.

Electric power is generated and transmitted in three phases for two significant reasons.
One, it facilitates smooth conversion of energy by applying uniform torque on gener-
ators and motors, unlike single phase AC which results in pulsating torque. This is an
engineering advantage as the rotors are well balanced. Two, it offers cost benefits as the
same amount of power can be transmitted with fewer conductors when compared to sin-
gle phase transmission. Currents and voltages add up to zero in a balanced three phase
system, eliminating the need for a common return wire2 . This is under the assumption
that all loads have equal impedance3 . However, in practice the impedances are slightly
different, hence requiring a return wire to complete the circuit. This is achieved by con-
necting the combined return wires to the common ground4 at both ends.

Transmission systems are modelled and analyzed as single phase AC systems for the
very reason of being balanced. That is, currents and voltages are equal in magnitude in
all three phases. Whereas in distribution systems, current and voltage magnitudes are
different because of the difference in loads and hence distribution systems are modelled
considering all three phases. In fact, it is interesting to note that residential and most
commercial circuits run on single phase AC and have two wires, one live and the other
neutral (return).
2
The three return wires are combined as a single return wire.
3
Impedance is the AC equivalent of resistance.
4
In electric circuits, ground refers to an electrically neutral reference that has 0 voltage. In this context,
it refers to the earth which acts as the return pathway of the circuit.

5
2.3 Consumption
Loads are devices that consume electric power and are characterized by impedance.
Theoretically, they can be broadly classified as resistive, inductive and capacitive loads.
Resistive loads are heating conductors that are seen in incandescent bulbs and heaters.
Inductive loads are all kinds of motors, fluorescent lamps and the transformers used
in power supplies. Capacitors generally do not do physical work like other loads but
are part of electrical circuits [4]. Based on usage, loads can be classified as residential,
commercial, industrial and electric railways. Another important category of loads is
consumer electronics [4].

From the system and also modelling perspective, electric power consumption is consid-
ered as aggregate load that combines several consumers. This may include households,
city blocks or entire cities and regions. Given that the electric power industry is largely
customer driven, capacity planning and serving instantaneous demand are very crucial
for grid operators. Load forecasting is a discipline in itself and plays an important role
in uninterrupted supply of electricity. For an interesting read about energy markets, we
refer to [6], section 3.6.

2.4 Challenges
The stochastic nature of renewable energy sources such as wind and sun poses unprece-
dented challenges to the electric grid. Solar and wind farms are highly uncertain in
generating power and cause severe problems to grid stability, possibly resulting in over-
loading and blackouts. Decentralized power generation by small windmills and rooftop
solar panels that are connected to the distribution system causes a change in power-flow
direction. This results in two-way traffic, making grid control and even power-flow
analysis a hard task. Electric vehicle charging is another difficult-to-predict scenario on
the distribution side of the electric grid.

The largest power outage in history occurred in north-east India on 30, 31 July, 2012
and affected 620 million people. In 2016, a blackout occurred in South Australia due to
storms that caused a cascading failure of the transmission system infrastructure, affect-
ing 1.7 million people. In May 2020, TenneT declared emergency state due to a high
voltage incident that occurred due to high RES infeed and low demands.

It is very likely that such events take place in the future owing to the rapid changes the
electricity systems are experiencing lately.

6
 Chapter 3

Modelling the Transmission System

Power flow analysis is the numerical analysis of the flow of electric power and involves
determining the operating state of the entire power system. The state describes how the
power system functions and helps to understand how the system responds to inputs. In
this chapter we derive the mathematical model of the transmission system and define
the power flow problem which is cardinal to power flow analysis.

3.1 Fundamentals of AC circuits

This section describes the characteristics of an AC circuit, as required to model the
transmission system. Definitions and equations are based on electric circuit theory.

3.1.1 Current and Voltage

In an AC system, current and voltage are sinusoidal functions characterized by ampli-
tude, frequency and phase. They are expressed as,

i(t) = Imax sin (ωt + ϕI ) v(t) = Vmax sin (ωt + ϕV ) (3.1)

where,

Imax = amplitude of current, A

Vmax = amplitude of voltage, V
ω = angular frequency, rad/s
t = time, s
ϕ = phase shift, rad

For load-flow calculations, average values of currents and voltages are preferred. Av-
eraging is done by considering Root Mean Sqaure (RMS) values of current and voltage

7
functions.
√ Since sinusoidal functions are perfectly symmetric, the effective or rms value
is 1/ 2 times the amplitude. Instantaneous current and voltage are now written as,

√ √
i(t) = 2|I| sin (ωt + ϕI ) v(t) = 2|V| sin (ωt + ϕV ) (3.2)

where |I| and |V| are rms values which are calculated as follows.

s s
∫T ∫T
1 1
|I| = i2 (t) dt |V| = v2 (t) dt (3.3)
T 0 T 0

where T = 2π/ω [s] is the period of sine wave. Intuitively, rms value is equal to the
DC equivalent that dissipates the same amount of electric power in a given resistor per
unit time.

In balanced three phase systems, current and voltage are equal in magnitude in all three
phases but shifted in phase by 2π/3 rad. Consequently, the power flow problem is
solved by considering only one phase and the other two phases are analyzed simply by
incorporating the phase shift. Note that it is convention to use the cosine function to
describe current and voltage in power flow analysis. For a balanced three phase system,
the governing equations are,

√ √
i(t) = 2|I| cos (ωt − ϕ − δ) v(t) = 2|V| cos (ωt − δ) (3.4)

where δ = {0, 2π/3, 4π/3} rad is the phase shift between the three phases and ϕ is the
phase shift between current and voltage.

3.1.2 Phasor notation

Steady-state power flow analysis can be considerably simplified by using phasors to
represent sinusoidal current and voltage functions. A phasor is an arrow that is imag-
ined to spin in the complex plane and it characterizes a sine wave by specifying its
magnitude and angle. Length of the phasor corresponds to amplitude or rms value, its
angle with respect to real axis corresponds to time and its rotation corresponds to angu-
lar frequency which is constant and is generally not considered for steady-state power
flow calculations. Considering sinusoidal current and voltage expressions:

i(t) = Imax cos (ωt − δI ) and v(t) = Vmax cos (ωt − δV ),

we use Euler’s identity ejϕ = cos ϕ + j sin ϕ and obtain,
√ √
i(t) = 2Re(|I|ejδI ejωt ) v(t) = 2Re(|V|ejδV ejωt )
√ √ (3.5)
= 2Re(Iejωt ) = 2Re(Vejωt )

8
where,
I = |I|ejδI and V = |V|ejδV

Here I and V are current and voltage phasors. In a balanced three phase system, current
and voltage values of one phase can be used to determine values of other phases just by
accordingly rotating the phasors.

3.1.3 Power
Considering the phase with δ = 0 and equations (3.4), instantaneous power can be
expressed as,

p(t) = v(t)i(t)
√ √
= 2|V| cos (ωt) 2|I| cos (ωt − ϕ)
(3.6)
= |V||I| cos ϕ[1 + cos (2ωt)] + |V||I| sin ϕ[sin (2ωt)]
= P[1 + cos (2ωt)] + Q[sin (2ωt)]
where P = |V||I| cos ϕ is called active power and Q = |V||I| sin ϕ is called reactive
power.

As evident from equation (3.6), instantaneous power is made up of two sinusoidal com-
ponents. The first component P[1 + cos (2ωt)] is unidirectional with average value P
and the second component Q[sin (2ωt)] is bidirectional with an average of 0.

Active power P is measured in watts [W]. It represents the power actually transmitted
or consumed by loads and is always positive. For instance, for purely resistive loads,
active power corresponds entirely to conversion of electric energy to heat or light. Ac-
tive power is also called real power or average power.

Reactive power Q is expressed in volt-ampere reactive [Var]. For loads with reactance,
phase difference between current and voltage ϕ is not zero and it results in instantaneous
power sometimes being negative which can be interpreted as power flowing backwards
from the load to the generator. This power that is oscillated back and forth through the
lines is exchanged between electric and magnetic fields and is not dissipated [4]. Reac-
tive power is also called imaginary power.

Power factor, often abbreviated as p.f. is defined by cos ϕ. When current lags voltage,
ϕ is positive and power factor is said to be lagging. When current leads voltage, ϕ is
negative and power factor is said to be leading. As ϕ varies from 0 to 90◦ , p.f. varies
from 1 to 0 corresponding to the loads from being purely resistive to being purely induc-
tive. A power factor of 1 is highly desired since lower power factors lead to increased
currents and higher heating losses in the power system [3].

9
The vector sum of P and Q is called complex power and is expressed as,

S = VI∗ = P + jQ (3.7)
where V and I are voltage and current phasors and I∗ is the complex conjugate of I.
Another important quantity is apparent power which is generally used to specify the
rating of an electrical apparatus. It is the product of current and voltage, regardless of
their phase shift. It is measured in volt-amperes [VA] and is written as,

S = |V||I| (3.8)

3.1.4 Impedance and Admittance

In AC circuits, the opposition to flow of current is called impedance which is the vector
sum of resistance and reactance. It is given by,

Z = R + jX (3.9)
where R is resistance (real part) and X is reactance (imaginary part). Impedance is mea-
sured in ohms [Ω] and comes with every device in an AC circuit. When X is positive,
reactance is inductive and jX = jωL where L is the inductance. When X is negative,
reactance is capacitive and jX = 1/jωC where C is the capacitance. Note that R, L and
C are always positive. When X is zero, impedance is purely resistive, indicating that
there are no inductors and capacitors in the circuit.

The inverse of impedance is called admittance denoted by Y. It is expressed as,

Y = 1/Z = G + jB (3.10)
where G is called conductance and B, susceptance. Considering the magnitudes of G
and B, admittance can be written as,
R X
Y= 2
−j 2 (3.11)
Z Z
where Z is the magnitude of Z. G,B and hence Y are measured in siemens [S].

Furthermore, Ohm’s law is extended to AC circuits as follows.

V = ZI or I = YV (3.12)

3.1.5 Kirchhoff’s Circuit Laws

Kirchhoff’s Voltage Law (KVL) states that the sum of voltages around any closed loop
in a circuit must be zero.
∑
Vj = 0 (3.13)
j

10
where Vj is the voltage across component j in the closed loop.

Kirchhoff’s Current Law (KCL) states that the currents entering and leaving any node
in the circuit must add up to zero.
∑
Iik = 0 (3.14)
k

where Iik is the current flowing from node i to node k.

Kirchhoff’s laws are extensively used to calculate currents and voltages in electrical
circuits.

3.1.6 Per unit system

It is common practice to normalize numerical values for ease of calculation. In per unit
(pu) system, the quantities of interest are expressed in terms of base value as follows.
actual value
per unit value =
base value
The per unit value is dimensionless. In power system analysis, voltages, currents, impedances
and powers are normalized. One of the advantages of using pu system is that the nu-
merical effect of an ideal transformer is reduced to that of a series impedance [7].

3.2 Network Topology

The transmission system is modelled as a network of nodes and edges. Nodes are called
buses and they represent points in the circuit where system components such as loads,
generators, tap transformers, phase shifters, shunts or substations are connected. Edges
are transmission lines that connect buses and hence system components to each other.
In this section, we look at the topology in detail and describe how the components are
modelled.

3.2.1 Buses and lines

Buses are considered to be electrically distinct, meaning that there exists an impedance
between them which sustains a potential difference. Each bus is characterized by the
following four quantities.

• Voltage phasor magnitude, V

• Voltage phase angle, δ

• Injected active power, P

• Injected reactive power, Q

11
Furthermore, buses are distinguished based on the parameters specified or controlled by
components that they are connected to. Following three types are commonly used in
power flow analysis.

Load Bus: As loads signify power consumption, they are modelled as such. At each
load bus, active power and reactive power are specified, which together constitute neg-
ative power injection. A load bus is called PQ bus, suggesting that P and Q are known.
V and δ are unknown, corresponding to the fact that loads do not control voltage.

Generator Bus: Generators are known to have control over active power and voltage.
Hence, a generator bus is referred to as PV bus. However, wind turbines do not have
voltage control and they are treated as PQ buses with a positive P. Another exception is
that some generators supply only active power and they are modelled as PQ buses with
Q = 0. The following reason clarifies this approach.

As mentioned earlier, supply and demand in an electric grid should always be in balance.
This is achieved by matching generation and consumption of active and reactive power.
Generators are solely responsible for active power balance, whereas reactive power can
also be balanced by using devices such as shunts1 . Hence, generators are modelled as
PV or PQ buses depending upon the parameters they control.

Slack Bus: The challenge with achieving power balance is that generation should also
account for transmission losses which are not known in advance. For active power, the
trick in practice is to make an empirical assumption of what the losses could be and get
a fixed power dispatch from all generators except one2 . This generator’s output is said
to be controllable. It takes up the slack by generating more power if losses are greater
than expected or less power if losses are smaller. Likewise, in power flow analysis, the
slack bus or swing bus is analogous to the variable generator. As real power balance
is a manifestation of steady frequency and hence of voltage angle, the phase angle δ is
specified in place of P for the slack bus. On the other hand, slack in reactive power
is shared by shunts and all generators that dispatch reactive power. Hence, V is spec-
ified for the slack bus as it is equivalent to requiring a balanced reactive power3 . Note
that it is convention to use δ = 0 for the slack bus. It acts as a global reference for timing.

Table 3.1 summarizes the above described distinction of buses and helps visualize the
parameters. N is the total number of buses in the network and Ng is the number of
generator buses.

Buses such as transmission substations that are not connected to generators or loads are
modelled as loads with P = Q = 0. A bus can also have both generator and load
connected. Such buses are modelled as PV buses with P = Pgen + Pload .
1
Shunts are devices that inject or consume reactive power.
2
Could also be a few but as done in literature, we consider them as one to substantiate a slack bus.
3
For a more detailed description about buses and choice of variables, we refer to [4], section 7.2.

12
 Table 3.1: Bus types and variables

Bus type Number of buses Known Unknown

Slack bus 1 δ, |V| P, Q
PV bus Ng P, |V| δ, Q
PQ bus N − Ng − 1 P, Q δ, |V|

A transmission line is modelled as an impedance between two nodes i and j. The

impedances of transmission lines are assumed to be time-invariant under any electric
potential and current. This allows the application of Ohm’s law to determine line cur-
rents. Since a balanced three phase system is modelled considering the single phase
equivalent, all transmission lines in the model correspond to one phase out of the three.

3.2.2 Tap Transformers and Shunts

The frequency of a transmission system, sometimes called system frequency, is constant
throughout the network whereas the voltage is not. The voltage largely depends on the
local situation of the system and as a consequence, can only be controlled locally [3].
Tap transformers and shunts are system components that play a very important role in
controlling the voltage across the network.

A tap transformer, also called tap-changing transformer is a transformer in which the

turns ratio4 can be adjusted. A mechanical tap is used to adjust the ratio. The voltages
on either side of the transformer are related as,

|V1 |
|V2 | =
t
where t is the turns ratio.

As there is a strong relation between reactive power exchange and voltage levels, shunts
are used to balance voltage levels by consuming or injecting reactive power. At a net-
work bus, reactive power consumption results in a lower bus voltage and reactive power
injection results in a higher bus voltage. Shunt capacitors inject reactive power whereas
shunt inductors consume reactive power. A shunt is modelled as a reactance zs = jxs
between the bus and the ground. The shunt admittance is defined as follows.
1 1
= −j = jbs
ys =
zs xs
For inductive shunts, xs is positive and for capacitive shunts, xs is negative. Note that
the shunt susceptance is bs = −1/xs .

4
Turns ratio is the ratio between number of windings on primary and secondary sides of a transformer.

13
3.3 The Power Flow Model
As the name implies, power flow or load flow simulations involve understanding the
flow of electric power from source to destination. Power flow gives insight about the
state of the transmission system and is one of the most important network computations.
State, also referred to as grid state, describes steady-state behavior of the network and
is defined by three quantities; power, current and voltage. Steady-state means that only
power frequency (50 or 60 Hz) is considered for calculations and the time step could
be minutes, hours, months or years5 . Given power injections at different parts of the
network, the objective is to compute voltage at every node and current in every line. We
derive power flow equations and define the power flow problem as follows.

At each node i of the network, complex power is defined by,

Si = Vi I∗i (3.15)
where Vi is the potential difference between the node i and ground, and Ii is the current
injected at node i. From Kirchhoff’s Current Law we have,

∑
N
Ii = Iik (3.16)
k=1

where Iik is the current between node i and node k ̸= i. That is, it’s the current flowing
from every node in the network to node i. From Ohm’s law, line current is related to
voltage as,

Iik = Yik Vk (3.17)

where Vk is the voltage at node6 k and Yik is the admittance7 of the transmission line
joining nodes i and k. In matrix form,

I = YV (3.18)
where I ∈ CN is the vector of current injections at nodes, V ∈ CN is the vector of node
voltages and Y = [Yik ] ∈ CN×N is called admittance matrix. The entries [Yik ] define
the line impedance between node i and node k. From (3.10) we have, Yik = Gik +jBik .
For nodes not directly connected to node i, Yik = 0 and hence Y is sparse and in KCL,
it is sufficient to sum only over nodes that are directly connected to node i.

5
Dynamic(kHz) and transient(MHz) analyses consider milliseconds and microseconds respectively
for calculations.
6
From Ohm’s law, Vk should have been the voltage drop across the impedance but we consider it as
voltage at node k for now, as we will see further that the potential difference between nodes i and k arises
in the power flow equations once we introduce phasors.
7
It is convenient to use admittance Y instead of impedance Z as we can define the admittance between
two unconnected nodes as 0.

14
Complex power at node i can now be written as,

Si = Vi (YV)∗i
∑N ∗
= Vi Yik Vk (3.19)
k=1

Using phasors and expanding Yik ,

∑
N
Si = |Vi ||Vk |ejδik (Gik − jBik )
k=1
∑N
= |Vi ||Vk | (cos δik + j sin δik ) (Gik − jBik ))
k=1

where δik = (δi − δk ) denotes the difference in phase angles between node i and k.
Considering the real and imaginary terms of complex power Si , we have the following
two equations for active and reactive power which are called power flow equations.

∑
N
Pi = |Vi ||Vk | (Gik cos δik + Bik sin δik ) (3.20a)
k=1
∑N
Qi = |Vi ||Vk | (Gik sin δik − Bik cos δik ) (3.20b)
k=1

The power flow problem, also called load flow problem can now be stated as:

Given the power injection S at each node,

find the voltage V at every node and current I in every line.

This problem is solved by computing V from the power flow equations and then com-
puting I using Ohm’s law and KCL.

The power flow equations form a system of non-linear equations for which a closed-
form solution is not known to exist. However, it is a root-finding problem and we use
well established methods such as the Newton Raphson iterative algorithm to find a nu-
merical solution.

Solving the power flow problem is of tremendous importance since it lies at the root of
various applications in power system analysis. In this report we focus on the following
applications which are relevant to the project.

15
1. Reactive power compensation: As explained in section 3.2.2, shunts are used to
regulate voltages across the transmission network. Reactive power compensation
assessment is made to quantify the amount of shunt inductors and capacitors banks
needed for voltage regulation in the future.

2. Year-round calculations: Year-round calculations involve evaluating power flow

for every hour of the year. The hourly data of supply and demand is fed into the
model and the voltages and the currents are analyzed. Year-round calculations
give a broad insight into the behavior of the transmission network over long pe-
riods of time. In power system planning, year-round calculations are carried out
for the coming decades to assess how the existing network needs to be expanded
in the future.

3. Contingency analysis: In electricity systems, contingency analysis is an investiga-

tion of scenarios wherein system components such as generators or transmission
lines are out of service or are taken down for maintenance. A standard criterion
in contingency analysis is the N-1 criterion which is often called N-1 secure (for
normal minus one). If an electric grid is called N-1 secure, it means that the grid
should be functional even if one system component such as a major transmission
line is out of service. For higher security, some electric grids are made N-2 secure
in which case the grid should be able to withstand two contingencies, that is, the
loss of two system components.

16
 Chapter 4

Power Flow Solvers

Digital solution methods to solve the power flow problem first appeared in the literature
in 1956 and major breakthroughs in power flow computations were made in the 1960s
[8]. There has been a lot of research in numerical methods to efficiently solve the power
flow problem ever since. In recent times, considering the rapidly growing sizes of elec-
tric grids, power flow solvers are of tremendous importance to power system operators.

Since the performance of a power flow solver depends largely on various factors such as
problem size, available computing power and ways of implementation, it is often very
hard to choose the right solver for a given problem. In this chapter, we describe some
of the most widely used power flow solvers.

4.1 Newton-Raphson
The Newton-Raphson (NR) method is a widely accepted root-finding algorithm that can
be used to solve a system of non-linear equations of the form F(x) = 0. Starting with
an initial approximation, the iterative scheme involves making successive corrections to
vector x. The correction vector ∆x is assumed to satisfy F(x + ∆x) = 0 at each iteration
and a first order Taylor expansion of F(x + ∆x) gives the NR iterative formula

−J(x)∆x = F(x) (4.1)

where J(x) is the Jacobian matrix calculated as Jik = ∂F∂xi (x) k
. The partial derivatives
represent the slopes of the tangent hyperplanes [8]. Algorithm (1) describes the basic
structure of the Newton-Raphson method. Traditionally, direct solvers are used to solve
the Jacobian matrix equation in each iteration. The residual norm ∥F(xk )∥ or the relative
k)
residual norm ∥ F(x
F(x0 )
∥ is used as a measure to check convergence. The Newton-Raphson
method is known to have quadratic convergence when iterates are close to the solution
[7]. The iteration process of the Newton-Raphson method for a one-dimensional func-
tion is shown in fig. 4.1.

17
 Algorithm 1: Newton-Raphson Method
k := 0
Initialize: x0
while not converged do
Solve for the correction: −J(xk )∆xk = F(xk )
Update the approximation: xk+1 = xk + ∆xk
k=k+1
end

F(x)

x
x2 x1 x0

Figure 4.1: Newton-Raphson iterations

The classical approach to initialize x0 is to use the flat start as initial value. That is, all
voltage angles are set to 0 and voltage magnitudes are set to 1 pu (equal to that of the
slack bus). For better convergence, the solutions of approximate methods such as DC
approximation (described in section 4.3) are used as initial values.

To solve the power flow problem using the NR method, F(x) can be formulated as power
or current mismatch functions. The unknown variable vector x can be represented in
three different coordinates such as polar, Cartesian and complex form as shown in table
4.1.

The two mismatch formulations of F(x) and three coordinate forms of x result in six
possible ways of applying the Newton-Raphson method to solve power flow problems.
These six methods are considered as the fundamental Newton power flow methods based
on which various modified versions are developed [9].

The most widely used version is power-mismatch formulation with polar coordinates
which is introduced in [10]. The current-mismatch versions with polar and Cartesian
coordinates developed in [9] are found to perform well for large scale transmission sys-

18
 Table 4.1: Variable x in different coordinates

Coordinates Variable x


Polar Vi = |Vi |ejδi [δ1 , . . . , δn , |V1 |, . . . , |Vn |]T
Cartesian (Vi = Vir + jVim ) [V1m , . . . , Vnm , V1r , . . . , Vnr ]T
Complex form (Vi ) [V1 , . . . , Vn ]T

tems. In this report, we review the three versions; polar power-mismatch version intro-
duced in [10], polar and Cartesian current-mismatch versions developed in [9]. For a
detailed comparison of NR methods, we refer to [9] where all six versions are investi-
gated with numerical experiments and a general framework for applying NR methods
to power flow problems in transmission and distribution systems is presented.

4.1.1 Polar power-mismatch version

The power-mismatch function F(x) is formulated as,

Fi (x) = ∆Si (x) = Ssp

i − Si (x)
∑
N
∗
= Ssp
i − Vi Yik Vk∗ (4.2)
k=1

where Ssp sp sp
i = Pi + jQi is the specified complex power injection at bus i and Si (x) is
the complex power computed at bus i which follows from (3.19).

In polar coordinates, ∆Si (x) is separated using (3.20) as,

∑
N
∆Pi (x) = Pisp − |Vi ||Vk | (Gik cos δik + Bik sin δik ) (4.3a)
k=1
∑N
∆Qi (x) = Qsp
i − |Vi ||Vk | (Gik sin δik − Bik cos δik ) (4.3b)
k=1

Using the polar power-mismatch function, the Newton-Raphson iterative formula can
be written as follows.
 11 12     
J J ∆δ ∆P
− = (4.4)
J21 J22 ∆|V| ∆Q
where the Jacobian sub-matrices are defined as J11 = ∂∆P
∂δ
, J12 = ∂∆P
∂|V|
, J21 = ∂∆Q
∂δ
,
∂Fi (x)
J22 = ∂∆Q
∂|V|
and the partial derivatives Jik = ∂xk
are calculated as follows.

19
 ∂∆Pi (x)
= −|Vi |(Gik cos δik + Bik sin δik )
∂|Vk |
∂∆Qi (x)
= −|Vi |(Gik sin δik − Bik cos δik )
∂|Vk |
(i ̸= k)
∂∆Pi (x)
= −|Vi ||Vk |(Gik sin δik − Bik cos δik )
∂δk
∂∆Qi (x)
= −|Vi ||Vk |(−Gik cos δik − Bik sin δik )
∂δk

!
∂∆Pi (x) ∑
= − 2|Vi |Gii + |Vk |(Gik cos δik + Bik sin δik )
∂|Vi | i̸=k
!
∂∆Qi (x) ∑
= − −2|Vi |Bii + |Vk |(Gik sin δik − Bik cos δik )
∂|Vi | i̸=k (i = k)
∂∆Pi (x) ∑
=− |Vi ||Vk |(−Gik sin δik + Bik cos δik )
∂δi i̸=k
∂∆Qi (x) ∑
=− |Vi ||Vk |(Gik cos δik + Bik sin δik )
∂δi i̸=k

To solve the linear system (4.4), it has to be modified based on the information available
at each bus for the following reason. We know from section 3.2.1 that at each PV bus,
P and |V| are specified whereas Q and δ are unknown. Hence, for each PV bus j, ∆Qj
cannot be formulated and the corresponding partial derivatives in the Jacobian matrix
i ∂∆Qi ∂∆Qj ∂∆Q
cannot be computed. As a result, we eliminate the entries ∂∆P ,
∂|Vj | δ|Vj |
, ∂|Vi | and ∂δi j
for all i = 1 . . . N from the Jacobian matrix J(x), ∆|Vj | from the correction vector ∆x
and ∆Qj from the power mismatch function F(x) for each PV bus j. Similarly, δ and
|V| are known for the slack bus and the corresponding entries in the linear system are
eliminated. The order of J(x) reduces to (2N − Ng − 2) and the vector x becomes,
 T
x = δ2 , . . . , δN , |VNg +2 |, . . . , |VN |
Note that conventionally, δ1 and |V1 | correspond to the slack bus. The modified linear
system is solved at each NR iteration.

4.1.2 Polar current-mismatch version

The current-mismatch function is formulated using the current equation (3.16) and the
complex power equation (3.15) as follows.

20
 Fi (x) = ∆Ii (x) = Isp
i − Ii (x)
 sp ∗ ∑ N
Si
= − Yik Vk (4.5)
Vi k=1
 sp ∗
Si
where Isp
i = Vi
is the complex current injection specified at bus i and Ii (x) is the
complex current computed at bus i.

The function ∆Ii (x) is separated into real ∆Iri (x) and imaginary ∆Im
i (x) current-mismatch
functions in polar form as follows.

Pisp cos δi + Qsp ∑ N

i sin δi
∆Iri (x) = − |Vk | (Gik cos δk − Bik sin δk ) (4.6a)
|Vi | k=1

Pisp sin δi − Qsp ∑ N

i cos δi
∆Im
i (x) = − |Vk | (Gik sin δk + Bik cos δk ) (4.6b)
|Vi | k=1

The NR iterative formula for the polar current-mismatch version can be written as,
 11 12     
J J ∆δ ∆Ir
− = (4.7)
J21 J22 ∆|V| ∆Im
r ∂∆Ir ∂∆Im ∂∆Im
where J11 = ∂∆I∂δ
, J12 = ∂|V|
, J21 = ∂δ
, and J22 = ∂|V|
. The partial derivatives are
calculated as follows.

∂∆Iri (x)
= −(Gik cos δk − Bik sin δk )
∂|Vk |
∂∆Imi (x)
= −(Gik sin δk + Bik cos δk )
∂|Vk |
(i ̸= k)
∂∆Iri (x)
= |Vk |(Gik sin δk + Bik cos δk )
∂δk
∂∆Imi (x)
= −|Vk |(Gik cos δk − Bik sin δk )
∂δk

∂∆Iri (x) Pisp cos δi + Qsp i sin δi

= − (Gii cos δi − Bii sin δi ) −
∂|Vi | |Vi | 2

∂∆Imi (x) P sin δi − Qsp

sp
i cos δi
= − (Gii sin δi + Bii cos δi ) − i
∂|Vi | |Vi |2
(i = k)
∂∆Iri (x) Pi sin δi − Qsp
sp
i cos δi
= |Vi | (Gii sin δi + Bii cos δi ) −
∂δi |Vi |
∂∆Imi (x) Pi cos δi + Qsp
sp
i sin δi
= −|Vi | (Gii cos δi − Bii sin δi ) +
∂δi |Vi |

21
Similar to power-mismatch version, the linear system (4.7) has to be modified. For a PQ
bus, computation of real and imaginary current-mismatch functions is straightforward
since the associated real and reactive power mismatches are known. Whereas represent-
ing a PV bus in the linear system is tricky and there are several approaches available in
literature. In this report, we review the new approach developed in [9] which is found
to be promising.

For each PV bus j, the reactive power Qj is considered as a dependent variable of |V|
and δ. The current-mismatch formulation is directly used. That is, ∆Iri (x) and ∆Im i (x)
sp
are calculated using Qi = Qj in equations (4.6) for each PV bus j at each iteration.
∂∆Ir ∂∆Im ∂∆Ir
The partial derivatives ∂|Vji| and ∂|Vji| in the Jacobian matrix J(x) are replaced by ∂Qji
∂∆Im
and i
∂Qj
for all i = 1 . . . N which are calculated as follows.

∂∆Iri (x)
=0
∂Qj
(i ̸= j)
∂∆Im
i (x)
=0
∂Qj

∂∆Irj (x) sin δj

=
∂Qj |Vj |sp
(i = j)
∂∆Irj (x) cos δj
=−
∂Qj |Vj |sp

The entries ∆|Vj | in the correction vector ∆x are replaced by ∆Qj for each PV bus j.
The initial reactive power Q0j is calculated for each PV bus j as follows.

∑
N
Q0j = |Vj ||Vk | (Gjk sin δjk − Bjk cos δjk )
k=1

The order of J(x) remains (2N − 2). At each NR iteration, the modified linear system
is solved and the reactive power Qj is updated using the computed correction ∆Qj .

4.1.3 Cartesian current-mismatch version

In Cartesian form, the current-mismatch function F(x) is separated as,

Pisp Vir + Qsp m ∑ N

i Vi
∆Iri (x) = − (Gik Vkr − Bik Vkm ) (4.8a)
(Vir )2 + (Vim )2 k=1

Psp V m − Qsp r ∑ N
i Vi
∆Im
i (x) = i r i2 − (Gik Vkm + Bik Vkr ) (4.8b)
(Vi ) + (Vim )2 k=1

22
and the Jacobian matrix equation is formulated as follows.
 11 12     
J J ∆V m ∆Ir
− = (4.9)
J21 J22 ∆V r ∆Im
r ∂∆Ir ∂∆Im ∂∆Im
where J11 = ∂∆I
∂V m
, J12 = ∂V r
, J21 = ∂V m
, and J22 = ∂V r
. The partial derivatives are
computed as,

∂∆Iri (x)
= −Gik
∂Vkr
∂∆Imi (x)
= Bik
∂Vkr
(i ̸= k)
∂∆Iri (x)
= Bik
∂Vkm
∂∆Imi (x)
= −Gik
∂Vkm

∂∆Iri (x) Pisp ((Vir )2 − (Vim )2 ) + 2Vir Vim Qsp

i
= −Gii −
∂Vir |Vi |4
∂∆Imi (x) Qsp ((Vir )2 − (Vim )2 ) − 2Vir Vim Pisp
= −Bii + i
∂Vir |Vi |4
(i = k)
∂∆Iri (x) sp
Qi ((Vir )2 − (Vim )2 ) − 2Vir Vim Pisp
= Bii +
∂Vim |Vi |4
∂∆Imi (x) Psp ((Vir )2 − (Vim )2 ) + 2Vir Vim Qsp
= −Gii + i i
∂Vim |Vi |4

In [9], the reactive power Qj is chosen as a dependent variable to represent a PV bus,

∂∆Ir ∂∆Im
similar to polar current-mismatch version. The partial derivatives ∂Qji and ∂Qij com-
puted as,

∂∆Iri (x)
=0
∂Qj
(i ̸= j)
∂∆Im
i (x)
=0
∂Qj

∂∆Irj (x) Vjm

= r 2
∂Qj (Vj ) + (Vjm )2
(i = j)
∂∆Irj (x) −Vjr
= r 2
∂Qj (Vj ) + (Vjm )2

23
are added to the Jacobian matrix J(x) and the correction ∆Qj is added to the correction
vector ∆x for each PV bus j. As a result, the Jacobian matrix becomes a rectangular
matrix. That is, J(x) ∈ R(2N)×(2N+Ng ) . In order to make the Jacobian matrix square,
the equation
Vr Vm
∆|V| = ∆V r + ∆V m
|V| |V|
is used with ∆|Vj | = 0 since |Vj | is specified for each PV bus j. This gives the relation,
Vjm
∆Vjr = − ∆Vjm
Vjr
which is used to add the column of the Jacobian matrix corresponding to the derivatives
∂∆Iri ∂∆Im ∂∆Iri ∂∆Im
∂V r and ∂V
i
r to the column corresponding to the derivatives ∂V m and i
∂V m
as follows.
j j j j

!
∂∆Iri ∂∆Iri Vjm ∂∆Iri
∆Vjm = − r ∆Vjm
∂Vjm ∂Vjm Vj ∂Vjr
!
∂∆Im
i ∂∆Im
i
Vjm ∂∆Im
i
∆Vjm = − ∆Vjm
∂Vjm ∂Vjm Vjr ∂Vjr
The correction vector ∆Vjr is now eliminated from the correction vector ∆x for each PV
bus j. The initial reactive power Q0j is calculated for each PV bus j as follows.
N 
∑ 
Q0j = Vjm (Gjk Vkr − Bjk Vkm ) − Vjr (Bjk Vkr + Gjk Vkm )
k=1
With the slack bus included, the order of J(x) remains (2N − 2). At each NR iteration,
the modified linear system is solved and the reactive power Qj is updated using the
computed correction ∆Qj .

Despite its widespread popularity, a drawback the Newton-Raphson method has is com-
putational complexity. The Jacobian matrix has to be computed in every iteration as it
depends on the current approximation of the solution. Which means, there is a new lin-
ear system (4.1) in every iteration for the algorithm to solve. This makes the solution
process computationally bound, particularly for applications such as contingency analy-
sis of large networks. These difficulties in solving the AC power flow problem have led
to extensive numerical studies and various simplified methods have been proposed and
used. The simplified methods involve making a series of approximations to the non-
linear powerflow problem (3.20). More the approximations made, the easier it is to find
a solution. However, note that the AC power flow methods such as Newton-Raphson
and all the approximate methods attempt to solve the same underlying power system.
In this report, we review two methods that are commonly found in literature: decou-
pled load flow and dc approximation. In situations where a full power flow model is an
absolute necessity, the solutions of these simplified methods are used as initial values,
essentially when it is quite certain that the flat start approximation doesn’t converge.

24
4.2 Fast Decoupled Load Flow
The Fast Decoupled Load Flow (FDLF) method is a simple and fast power flow solu-
tion technique which is derived from Newton’s method. We briefly describe the basic
version of the formulation here and refer to [3, 11] for further details.

The polar power-mismatch version of the Newton-Raphson method described in section

4.1.1 is used as a starting point to derive the decoupled load flow formulation. The linear
system that is solved in every iteration of the NR polar power-mismatch version is given
by (4.1) as,
 11 12     
J J ∆δ ∆P
− = (4.10)
J21 J22 ∆|V| ∆Q
The decoupling principle involves making the following assumptions which generally
hold for power systems under normal operating conditions [3].

1. Voltage angle differences between buses are small.

cos δik ≈ 1; sin δik ≈ δik

2. Transmission line susceptances are much larger than conductances.

Gik sin δik ≪ Bik

3. The reactive power injected into a node is much smaller than the reactive flow that
would result if all lines connected to that bus were short circuited to reference [3].

Qi ≪ Bii |Vi |2

Evaluating the partial derivatives J12 = ∂∆P

∂|V|
and J21 = ∂∆Q∂δ
(see section 4.1.1 for
equations) with the first two assumptions shows that the Jacobian sub-matrices J12 and
J21 are small and can be neglected. Additionally, the terms [J22 ik ] are multiplied with
voltage magnitude |Vi |, the convenience of which will be clear in the derivation below.
The following decoupled system of equations is then obtained.

    
− J11 ∆δ = ∆P (4.11a)

h ih i  
− eJ22 e = ∆Q
∆|V| (4.11b)

∂∆Pi e22 e ∆|Vi |

where, J11ik = ∂δk , Jik = |Vi | ∂|Vk | and ∆|Vi | = |Vi | . The Jacobian sub-matrices J
∂∆Qi 11

and eJ22 are computed using the partial derivatives defined in section 4.1.1 with the three
assumptions stated above as follows.

25
 ∂∆Pi ∂∆Qi
= |Vi | = −|Vi ||Vk | (Gik sin δik − Bik cos δik ) (i ̸= k)
∂δk ∂|Vk |
≈ |Vi ||Vk |Bik

∂∆Pi ∑
= |Vi ||Vk |(Gik sin δik − Bik cos δik ) (i = k)
∂δi i̸=k

∑
N
= Bii |Vi | + 2
|Vi ||Vk | (Gik sin δik − Bik cos δik )
k=1
= Bii |Vi | + Qi 2

≈ Bii |Vi |2

∂∆Qi ∑
|Vi | = 2Bii |Vi |2 − |Vi ||Vk |(Gik sin δik − Bik cos δik ) (i = k)
∂|Vi | i̸=k

= Bii |Vi |2 − Qi
≈ Bii |Vi |2

The decoupled system (4.11) is now written as,

    
− |V|T B|V| ∆δ = ∆P

 h i  
− |V| B|V| T e
∆|V| = ∆Q

where B = [Bik ] ∈ RN×N is the matrix of line susceptances. Finally, the terms |Vi |T
are taken to the right hand side and the terms |Vi | are set to 1 pu. A distinction is made
at this stage for matrix B in the two linear systems for convenience.

With all the above modifications, the final decoupled power flow equations become,

   h i
− B ′
∆δ = e
∆P (4.12a)

   h i
− B ′′
∆|V| = e
∆Q (4.12b)

where ∆P ei = ∆Pi and ∆Q e i = ∆Qi . The decoupled system (4.12) is modified to rep-
|Vi | |Vi |
resent a PV bus, similar to NR polar power-mismatch version as described in section
4.1.1. The order of the systems (4.12) will then be (N−1) and (N−Ng −1) respectively.

26
The matrices B′ and B′′ depend only on network parameters and are constant in every
iteration. This means that the matrices have to be calculated and factorized only once,
leading to faster computations of the power flow problem even though the number of
iterations needed for convergence could be higher because of the approximations made.
The steps involved in the fast decoupled load flow method are given in the following
algorithm.

Algorithm 2: Decoupled load flow

k := 0
Initialize: δ0 and |V|0
while not converged do
Compute: ∆P e
Solve for the correction ∆δ: − [B′ ] [∆δ] = [∆P] e
k+1 k k
Update the approximation: δ = δ + ∆δ
Use δ k+1
to compute ∆Q e
e
Solve for the correction ∆|V|: − [B′′ ] [∆|V|] = [∆Q]
Update the approximation: |V| k+1
= |V| + ∆|V|
k k

k=k+1
end

4.3 DC approximation
DC approximation or DC load flow is a linear approximation of the power flow prob-
lem. The extent to which the non-linear power flow equations (3.20) are approximated
and the kinds of assumptions made may vary based on the problem or the application.
In fact, ’DC’ refers to the collection of approximations made such that the network is
decoupled. In some cases, the FDLF method is also considered as a DC approximation
method. However, there is a fundamental difference between FDLF and DC approx-
imation methods. In FDLF, the non-linear system is solved in each iteration and the
approximation is made only to the Jacobian matrix. In DC load flow methods, the non-
linear equations are linearized to speed up computation, which considerably affects the
accuracy of the final solution.

The DC load flow method described in [3] is derived as follows. The following approx-
imations are made to the power flow problem (3.20).

• Bus voltage magnitudes are set to 1 pu in active power equations (3.20a).

• Voltage magnitudes are approximated as: |V| = 1 + ∆|V| and 1

1+∆|V|
= 1 − ∆|V|
in reactive power equations (3.20b).

• Conductances of transmission lines are neglected: Gik ≪ Bik .

• Voltage angles are small: cos δik ≈ 1 and sin δik ≈ δik .

27
Under these assumptions, the linearized version of the power flow problem (3.20) is
given by,

∑
Pi = Bik δik (4.13a)
k̸=i
∑
Qi + Bii = (Qi − Bii )∆|Vi | + (1 + ∆|Vk |)Bik (4.13b)
k̸=i

where ∆|V| represents the deviation of the voltage magnitude from 1 pu voltage level.

The DC approach provides a fairly good approximation of voltage magnitudes and an-
gles which can be used as initial values in NR or FDLF methods. It is also claimed in
[12] that while the DC approximation leads to some loss of accuracy, the results match
fairly closely with full power flow solution. The following approach of DC approxima-
tion is presented in [12] as the most dramatic of DC approximation methods.

The assumptions made to the non-linear power flow problem are:

• the reactive power balance equations are ignored,

• voltage magnitudes are set to 1 pu,

• line losses are ignored.

These assumptions reduce the non-linear power flow problem to the system of linear
equations:

[B′ ] [δ] = [P] (4.14)

where [B′ ] is the line susceptance matrix, [δ] is the vector of bus voltage angles and [P]
is the active power injection vector.

It should be noted that the lack of losses in the DC solution can be reasonably com-
pensated for by increasing the total load by the amount of estimated losses. The DC
approach (4.14) has the following significant advantages over the Newton-Raphson
method.

1. The linear system is half the size of the full problem since only the active power
mismatch equations are solved.

2. The algorithm is not iterative, requiring just one single solution of (4.14).

3. The matrix B′ is dependent only on network parameters and needs to be factorized

only once.

We refer to [12] for further details, where a comparison between AC and DC methods
is also made.

28
4.4 Gauss-Seidel
Gauss-Seidel method is another iterative technique that can be used to solve the non-
linear power flow problem. The iterative scheme is derived from the complex power
equation (3.19) with complex voltage Vi as the iteration variable.

Si = Vi (YV)∗i
!∗
∑
N
= Vi Yik Vk
k=1
!∗
∑
∗ ∗
= Vi Yik Vk + Vi Yii Vi ⇐⇒
k̸=i
!∗
∑
∗ ∗
Vi Yii Vi = Si − Vi Yik Vk ⇐⇒
k̸=i
!
1 Si ∑ ∗ ∗
Vi∗ = ∗ − Y V ⇐⇒
Yii Vi k̸=i ik k
!
1 S∗i ∑
Vi = − Yik Vk
Yii Vi∗ k̸=i
!
1 Pi − jQi ∑
= − Yik Vk (4.15)
Yii Vi∗ k̸=i

The fixed point equation (4.15) leads to the following iterative formula.
!
1 Pi − jQ i
∑
Vih+1 = − Yik Vkh , h = 0, 1, 2, . . . (4.16)
Yii Vih∗ k̸=i

where Vi0 is a given initial approximation for each bus i. Equation (4.16) is evaluated at
each iteration until convergence is met. If the approximations Vih are computed at once
for all buses and applied at once in the next iteration, the iterative procedure is called
Jacobi iteration. If the approximations are computed for one bus at a time and immedi-
ately used in the same iteration, the procedure is referred to as Gauss-Seidel iteration.

For a PQ bus, it is straightforward to apply the iterative formula (4.16), whereas for a
PV bus, modifications are required, similar to Newton-Raphson method.

The Gauss-Seidel method is flexible and relatively easy to implement but it is gener-
ally slow and inefficient for large systems. The iterative scheme has a tendency to use
a lot of computations, particularly in large scale problems, and could also converge to
incorrect solutions. Hence, the Gauss-Seidel method is not preferred in practice. How-
ever, despite the shortcomings, it can be used to get a good perspective on power flow
problems [13].

29
4.5 Summary
The following can be inferred from the power flow solvers that are described in the
preceding sections (4.1 to 4.4).

• The Newton-Raphson method is widely regarded as the most commonly pre-

ferred power flow solver [8]. Several formulations of the NR method are found
in literature. Among the six fundamental NR methods, the most widely used ver-
sion is the power-mismatch formulation with polar coordinates [10]. The current-
mismatch version with polar and Cartesian coordinates are found to give the best
results for transmission networks [9].

• The Gauss-Seidel method is another full power flow solver that is used in power
flow analysis. Despite the ease of implementation that the Gauss-Seidel method
offers, it is seldom preferred in practice due to computational intensity [13].

• The FDLF method offers computational benefits against full power flow solvers
by making approximations as described in section 4.2. The FDLF method is ad-
vantageous in terms of speed and simplicity compared to full power flow solvers
[11]. However, the accuracy of the solution is not on par with full power flow
solvers because of the approximations made.

• The DC approximation method is the simplest of approximate methods to solve

the power flow problem. It has been concluded in [12] that the results of the DC
approximation method match fairly closely with full power flow solvers despite
the loss in accuracy. However, the DC approach considers only active power
equations and hence can only substitute a full power flow solver for applications
that do not require reactive power to be taken into account.

We would like to emphasize that the selection of a power flow solver is largely problem
specific. For instance, while the DC approximation method is well suited for problems
that are not sensitive to the approximations made, it cannot be used for applications that
require calculation of reactive power in the network. Full power flow solvers such as
the Newton-Raphson method and its formulations are always necessary for AC power
flow analysis. Hence, for reactive power compensation assessment, AC power flow
simulations are absolutely necessary whereas for capacity planning, the DC approach
could result in a fair approximation.

In the remainder of this section, we investigate the convergence properties of the Newton-
Raphson method and evaluate the differences between AC and DC approaches of solv-
ing the power flow problem.

30
Power flow convergence
Mathematically, we call a problem well-posed if the following conditions hold.

1. Existence: There exists at least one solution.

2. Uniqueness: There is at most one solution.

3. Stability: The solution depends continuously on input data.

A problem which is not well-posed is called ill-posed. Figure 4.2 shows ill-posedness
of three one dimensional functions. In power flow analysis, the power flow problem
could be ill-posed due to any of the following reasons.

• The system might not have a solution. For example, in fig. 4.2a, f(x) never
crosses the x - axis, hence it has no solution.

• Even though a solution exists, the power flow solver cannot find it or convergence
is very difficult. For example, fig. 4.2b shows the tendency to diverge when the
initial value x1 is far from the solution. It is clear that x3 > x1 and continuing the
iteration further from x3 causes the solution to diverge.

• The system might have multiple solutions. For example, in fig. 4.2c, f(x) crosses
the x - axis twice, leading to two solutions. For power flow problems, the practical
consequence of this is that the solution may converge to a lower voltage value.

Many formulations of the power flow solution methods are available in the literature to
fix ill-posedness of power flow problems. These formulations are found to be highly
efficient in ensuring convergence and they also help to determine the existence of a so-
lution. In this report we describe the optimal multiplier method as presented in [14] and
the line search method of finding the optimal multipliers as explained in [15].

Optimal multiplier method: The Newton-Raphson method has local quadratic con-
vergence as mentioned in section 4.1. That is, the NR method converges quadratically
when the initial value is close to the solution. A method that converges for any ini-
tial value is called globally convergent. The optimal multiplier method ensures global
convergence and aims to solve the problems of false divergence and convergence to in-
correct solutions of the NR method. The optimal multiplier method is popularly known
as the Iwamoto multiplier method. The core idea is that a scalar multiplier µ is intro-
duced in the update step xk+1 = xk + ∆xk of the NR method (see algorithm 1) as
follows.

xk+1 = xk + µ∆xk
Many studies have been carried out in order to determine the optimal multiplier µ∗ . In
[14], µ∗ is determined as follows.

31
 4
1
3
x2
f(x)

2 0 x1
x3
1
−1
0
−2 −1 0 1 2 −2 −1 0 1 2
x x
(a) No solution (b) Divergence

solutions
1
f(x)

−1

−4 −3 −2 −1 0 1 2
x
(c) Multiple solutions

Figure 4.2: Ill-posedness

In the Newton-Raphson method, the first order Taylor expansion of the mismatch func-
tion F(x) is expressed as (see section 4.1),

F(x + ∆x) = F(x) + J(x)∆x = 0

where the higher order terms of the expansion are neglected. In [14], the higher order
terms are collectively expressed as the mismatch function evaluated at the correction
∆x. That is,

F(x + ∆x) = F(x) + J(x)∆x + F(∆x) = 0

Incorporating the multiplier µ,

F(x + µ∆x) = F(x) + µJ(x)∆x + µ2 F(∆x) = 0

A cost function C : Rn → R which represents the proximity of the approximation to
the solution defined as,
1
C(x) = F(x)T F(x) (4.17)
2

32
is used to determine the optimal value of µ by evaluating the following expression.
dC
=0
dµ
Algorithm 3 describes the modified Newton-Raphson algorithm with the optimal mul-
tiplier incorporated.

Algorithm 3: NR algorithm with optimal multiplier

k := 0
Initialize: x0
while not converged do
Solve for the correction: −J(xk )∆xk = F(xk )
Compute µk by evaluating dC dµ
=0 •
k
Set ∆x = µ ∆xk k
•
Update the approximation: xk+1 = xk + ∆xk
k=k+1
end

Line search technique: To determine the optimal multiplier µ, the line search method
is used in [15]. The line search is an iterative method to find a local minimum x∗ of an
objective function f : Rn → R. In our case, at each iteration the following optimization
problem is solved to determine µ∗ .

min C(xk + µ∆xk )

µ∈[0,1]

Since C(x) represents the proximity of the approximation to the solution, the scalar mul-
tiplier µ must satisfy C(xk + µ∆xk ) < C(xk ) at each iteration. Hence, the search is
limited to µ ∈ [0, 1]. The line search procedure is described as follows.

The procedure is similar to the binary search algorithm and starts with [µ1 , µ2 ] = [0, 1].
The domain is then divided in half. That is, [µ1 , µ3 , µ2 ] = [0, 0.5, 1]. The function C(x)
is evaluated at the three points and the following conditions are checked.

C(µ1 ) > C(µ3 )

C(µ2 ) > C(µ3 )

• If the conditions hold true, the two subdomains [0, 0.5] and [0.5, 1] are further
divided in half and the procedure is continued with [µ4 , µ3 , µ5 ] = [0.25, 0.5, 0.75]
and so on (see fig. 4.3a).

• If one of the conditions is violated, the corresponding subdomain is excluded and

the procedure is continued. That is, if C(µ2 ) < C(µ3 ), the three points will be
[µ3 , µ4 , µ2 ] where µ4 is the point at the center of [µ3 , µ2 ] (see fig. 4.3b).

33
 • The procedure terminates when the search domain converges to a point, which is
the optimal multiplier µ∗ .

2.5 2.5

2 2

1.5 1.5
C(µ)

1 1

0.5 0.5

0 0
µ1 µ4 µ3 µ5 µ2 µ1 µ3 µ4 µ2
µ µ
(a) (b)

Figure 4.3: Line search: subdividing the search domains

AC vs DC approach
The benefits of the DC approximation method such as simplicity and computational
advantage over full power flow methods are well justified only when the assumptions
made in the DC approach are realistic. In [16], the assumptions of the DC power flow
method are analyzed and quantified using indices. Here we describe the main ideas in
brief.

The assumptions made in the DC approximation method and their respective decisive
factors that can be used to judge the accuracy of the DC solution are described as follows.

1. Small voltage angles: cos δik ≈ 1 and sin δik ≈ δik .

These assumptions are often said to be useful only for weakly loaded networks.
However, if the voltage angles are actually small, this approximation does not
lead to significant errors. This is confirmed in [16] with an experiment conducted
on the Belgian high voltage grid consisting of about 900 lines and 700 buses with
voltages from 70 kV to 380 kV and a peak load of 13 GW. The voltage angle
differences were found to be below 7◦ . In 94% of the lines, the voltage angle
differences were found to be lower than 2◦ . Fig. 4.4 shows the relative error in
active power as a consequence of the assumptions made to voltage angles. The
relative error is calculated as follows.

Pac − Pdc
Perror = × 100
Pac

34
 Figure 4.4: Effects of voltage angle approximations

2. Negligible line resistance: R≪X

In actual electricity networks, it is almost impossible to neglect line resistances.
In order to understand the consequences of this assumption, the ratio of reactance
to resistance X/R is used as a measure. In [16], it has been shown that an X/R
ratio higher than 4 guarantees that the error in active power will be lesser than
5%.

3. Flat voltage profile: All voltage magnitudes are set to 1 pu.

The deviations of voltages from the assumed value (1 pu) is considered as the
decisive factor to assess the effects of the assumption. The voltage deviations are
measured by means of standard deviation using the following expression.
v
u
u 1 ∑
n
SV = t (Vi − V)2
n − 1 i=1

The Belgian high voltage network mentioned earlier is taken as an example in

[16] to check the likelihood of a favorable voltage profile. It has been found that
Perror is very sensitive to voltage deviations. Examples of voltages in actual
power systems have shown that the assumption of a flat voltage profile is the
most critical one and it is the biggest source of Perror . Hence, the flat voltage
profile is considered to be of high importance for the solution accuracy of the DC
approximation method. Flat voltage profile means that the standard deviation of
voltages is lesser than 0.01.

It is to be noted that the analyses of the assumptions 1 and 3 made in [16] are case
specific and can be considered for reference only when the actual phase angles and
voltages of the network are known. Whereas the decisive factor of assumption 2 can be
used to check the feasibility of the DC load flow method in all cases since resistances
and reactances are characteristics of the network and are always known.

35
 Chapter 5

Software Packages

Given the importance of digital solution methods to power flow analysis, there are many
commercial software packages available in the market. These software packages are
quite powerful and they are extensively used in the industry for power system operation,
control and planning among many other applications. In this chapter we explore Pow-
erFactory, PSS® E and pandapower and understand their capability to solve the power
flow problem.

5.1 PowerFactory
PowerFactory is an engineering tool for the analysis of transmission, distribution and
industrial electrical power systems. It is designed as an advanced integrated and inter-
active software package dedicated to electrical power systems in order to achieve the
main objectives of planning and operation optimization. PowerFactory has a single-line
graphical interface which includes drawing functions, editing capabilities, and static
and dynamic calculation features. The simulation functions offered by PowerFactory
include power flow analysis, contingency analysis, optimal power flow among many
other functions. In this report, we focus on power flow analysis capabilities and refer to
[17] for further details. PowerFactory is licensed by DIgSILENT GmbH.

PowerFactory offers both AC and DC load power methods. In AC power flow method,
the user can select one of the following formulations for solving the power flow problem.

1. Newton-Raphson power mismatch

2. Newton-Raphson current mismatch

PowerFactory allows the calculation of both balanced and unbalanced power flows. It
is claimed in [17] that the Newton-Raphson power mismatch version converges best for
large transmission systems, especially when heavily loaded and the Newton-Raphson
current mismatch version converges best for unbalanced distribution systems. How-
ever, as we saw in section 4.1, the current-mismatch formulation can also be used for

36
transmission systems.

In DC power flow method, only active power flow without losses is considered as ex-
plained in section 4.3 with the linear system (4.14).

PowerFactory has options for integration with Python scripts. Python scripts are gener-
ally used with PowerFactory for automation of tasks, creation of user defined calculation
commands and integration of PowerFactory into other applications. PowerFactory also
supports interfaces for softwares such as PSS® E and MATLAB.

5.2 PSS®E
PSS® E is a power system simulation and analysis tool for power transmission operation
and planning. Similar to PowerFactory, it offers a wide range of simulation functions.
For AC power flow calculations, PSS® E has the following iterative schemes along with
a few other modified methods available.

1. Gauss-Seidel

2. Newton-Raphson

3. Decoupled Newton-Raphson

Since the convergence properties of solvers depend on the network and load attributes,
the following procedure is suggested in [18] to solve the power flow problem, particu-
larly in situations where the characteristics of new power flow problems are not known.

1. Use flat start values as initial values.

2. Execute Gauss-Seidel method until the corrections made to voltages and angles
decrease to, for instance, 0.01 or 0.005 pu in consecutive iterations.

3. Switch to Newton-Raphson method and execute it until the problem converges or

until there are signs of failure to converge.

4. If Newton-Raphson method does not converge within 8 to 10 iterations, switch

back to Gauss-Seidel.

PSS® E also has an embedded Python interpreter which can be used to access and run
models in PSS® E from Python scripts.

37
5.3 pandapower
pandapower is an open source tool for power system modelling, analysis and optimiza-
tion with a high degree of automation. pandapower combines the data analysis library
pandas and the power flow solver PYPOWER to create a network calculation program.
PYPOWER is a port of MATPOWER to the Python programming language. MAT-
POWER offers the following power flow methods and a few other modified ones.

1. Newton-Raphson
2. Newton-Rapshon with Iwamoto multiplier
3. Gauss-Seidel
4. Fast decoupled load flow
5. DC power flow

MATPOWER includes four algorithms for solving the AC power flow problem which
correspond to the four fundamental Newton-Raphson formulations: power and current
mismatch formulations with polar and cartesian coordinates. The implementation of
these formulations in MATPOWER is based on [9]. The default solver uses the power
mismatch formulation with polar coordinates of the Newton-Raphson method (see sec-
tion 4.1.1). It is mentioned in [19] that the Gauss-Seidel method is included only for
academic interests as it has many disadvantages compared to the Newton’s method.

By default, the AC power flow solvers in MATPOWER solve the power flow problem
without considering voltage limits, line current limits and generator limits. Currently,
none of the solvers include options for automatic updating of transformer taps and for
satisfying constraints such as voltage limits [19]. However, there is an option to keep
the generator reactive power within limits, but at the expense of voltage setpoints. That
is, when the generator reactive power is kept within limits to ensure convergence, the
voltages could go beyond the safe operating range which is generally set to 0.9 to 1.1
pu. This is based on a brute force technique which adds an outer loop around the AC
power flow solver. If the reactive power limit of a generator is violated, the reactive
power injection by that generator is force fixed at the limit and the power flow is solved
again. This procedure is repeated until there are no more violations.

MATPOWER offers many benchmark grids to evaluate power flow algorithms. For our
project, we consider the following test cases which represent parts of the European high
voltage transmission network.

• Case 1354pegase: 1354 buses, 1751 lines.

• Case 2869pegase: 2869 buses, 4051 lines.
• Case 9241pegase: 9241 buses, 13797 lines.

38
 Chapter 6

Conclusion

We draw the following conclusions from the literature review.

The power flow problem is the problem of finding the voltage magnitudes and phase
angles in every bus of the network. The voltage phasors are then used to determine the
line currents and thus the power flow in every line of the network. The power flow
problem is a non-linear system of equations for which a closed-form solution does not
exist. However, a numerical solution can be found by using iterative algorithms such as
the Newton-Raphson method.

The Newton-Raphson method is widely regarded as the most distinguished full power
flow solver. It is a root finding iterative algorithm that is used to solve a system of
non-linear equations of the form F(x) = 0. Among several formulations of the NR
method found in the literature, the current mismatch version with polar and Cartesian
coordinates are found to perform the best for transmission networks [9]. The classi-
cal Newton-Raphson method has local quadratic convergence. Meaning, the method
has quadratic convergence when the initial value is close to the solution. However, to
achieve global convergence, that is to achieve convergence for any initial value, further
modifications of the method are necessary. Furthermore, when the problem is ill-posed,
the classical NR method tends to diverge.

The optimal multiplier method is commonly used to solve ill-posed power flow prob-
lems. That is, the optimal multiplier method helps to avoid false divergence and con-
vergence to incorrect solutions, and also ensures global convergence. Using the optimal
multiplier method, the existence of the solution from an initial value can be easily judged
and if a solution exists, the power flow solution never diverges [14]. The optimal mul-
tiplier method is also known as Iwamoto multiplier method. To determine the optimal
multiplier in every iteration of the NR method, the linesearch is an efficient technique
that is commonly used [15].

39
The DC power flow method can be considered feasible to solve the power flow problem
if the ratio of reactance to resistance X/R is greater than 4 [16].

We could find little material in the literature about automating power flow simulations.
The closest we have found is a subroutine in MATPOWER that keeps generator reac-
tive power within limits at the expense of voltage setpoints. This is not a mathematical
approach but a brute force technique [19].

The software PowerFactory has a Python Application Programming Interface (API) that
makes the functionality of PowerFactory accessible from a Python script. The API can
be used to develop an interface in Python which can access the grid models and run
simulations in PowerFactory. This functionality is currently used in the industry and
academia to automate certain calculations in PowerFactory. However, the API is meant
more to make the PowerFactory functionalities better accessible than to extract the grid
models from PowerFactory and solve them on a different solver such as pandapower for
the purpose of research.

Research Questions
In this project we focus on two main research questions: automating AC power flow
simulations and comparison of AC and DC power flow methods. The two research
objectives correspond to two applications respectively: reactive power compensation
assessment and capacity planning.

1. Automating AC power flow simulations: The goal is to develop an automated

solver which can perform AC year-round simulations for the European high volt-
age transmission network model, in order to perform reactive power compensa-
tion calculations. We take the following steps in order to develop the automated
model.

• The European transmission network model is found to be the one which has
convergence issues the most. However, the exact reason for divergence is
not clear yet. We investigate the European grid model in detail and find the
possible reasons for the solver to diverge. We start with the fact that even
though the European grid model converges for a particular set of inputs, it
is found to diverge when the inputs change.
• We use the Newton-Raphson formulations described in section 4.1 to de-
velop the automated solver for the European grid model. This involves de-
veloping mathematical methods such as the optimal multiplier method that
ensure convergence and also coming up with techniques to keep the voltages
across the transmission network within limits, run-time, during year-round
simulations.

40
 • To achieve the above objectives, the bottleneck is that the European grid
model is built in PowerFactory, the solvers of which cannot be accessed and
modified. We try to use the Python API to develop an interface which can
translate the grid models of PowerFactory to grid models that can be used
in pandapower (see fig. 6.1).
• We start the implementation of algorithms in pandapower with test networks
that are described in the next section.

Figure 6.1: PowerFactory to pandapower

2. AC vs DC approach: The solutions of the AC power flow simulation models

will be compared with the DC solutions. The consensus at the moment is that a
difference in results higher than 5% makes AC power flow simulations justifiable
for capacity planning. If the AC power flow method is found to be better than the
DC approach, we make a cost analysis of the AC power flow method for doing
year-round simulations. We start the comparison with the decisive factors that are
described in section 4.5 (AC vs DC approach), which give a framework for the
DC solutions to be considered feasible.

Transmission Networks
We consider the following transmission networks from TenneT for our experiments in
this project.

1. 220 and 380 kV Dutch grid: 465 buses.

2. 110, 150, 220 and 380 kV Dutch grid: 1500 buses (see fig. 6.2).

3. The European grid which includes the 220, 380 kV Dutch grid connected to Ger-
many, Belgium, Luxemburg and France: 12400 buses.

41
 Figure 6.2: Grid map Netherlands

To test the algorithms, the test networks shown in table 6.1 that are available in pan-
dapower will be considered.

Table 6.1: Test networks in pandapower

Test case Buses Lines Loads Shunts Generators

1354pegase 1354 1751 621 1082 259
2869pegase 2869 4051 1311 2197 509
9241pegase 9241 13797 4461 7327 1444

42
 Bibliography

[1] Carl Sulzberger. Pearl street in miniature: Models of the electric generating station
[history]. IEEE Power and Energy Magazine, 11(2):76–85, 2013.

[2] TenneT. Electricity Producers: The Netherlands and Germany.

[3] Pieter Schavemaker and Lou van der Sluis. Electrical Power System Essentials.
John Wiley & Sons, Inc., Sussex, United Kingdom, 2008.

[4] Alexandra von Meier. Electric Power Systems. John Wiley & Sons, Inc., Hoboken,
NJ, USA, 2006.

[5] M.E. Kootte, J.E. Romate, and C. Vuik. Load Flow Computations for (Integrated)
Transmission and Distribution Systems. A Literature Review. Technical report,
Delft University of Technology, 2020.

[6] Fons van der Plas. Power Grid Failures. Thesis bsc mathematics, Radboud Uni-
versity Nijmegen, 2019.

[7] R. Idema, D.J.P. Lahaye, and C. Vuik. Load Flow Literature Survey. Technical
report, Delft University of Technology, 2009.

[8] Brian Stott. Review of Load-Flow Calculation Methods. Proceedings of the IEEE,
62(7):916–929, 1974.

[9] Baljinnyam Sereeter, Cornelis Vuik, and Cees Witteveen. On a comparison of

Newton–Raphson solvers for power flow problems. Journal of Computational
and Applied Mathematics, 360:157–169, 2019.

[10] William. F. Tinney and Clifford E. Hart. Power Flow Solution by Newton’s
Method. IEEE Transactions on Power Apparatus and Systems, PAS-86(11):1449–
1460, 1967.

[11] B. Stott and O. Alsac. Fast decoupled load flow. IEEE Transactions on Power
Apparatus and Systems, PAS-93(3):859–869, 1974.

43
[12] Thomas J. Overbye, Xu Cheng, and Yan Sun. A comparison of the AC and DC
power flow models for LMP calculations. In Proceedings of the Hawaii Interna-
tional Conference on System Sciences, volume 37, pages 725–734, 2004.

[13] James L. Kirtley. Electric power principles: sources, conversion, distribution, and
use. John Wiley & Sons, Ltd, 2010.

[14] S. Iwamoto and Y. Tamura. A load flow calculation method for ill-conditioned
power systems. IEEE Transactions on Power Apparatus and Systems, PAS-
100(4):1736–1743, 1981.

[15] Patricia Rousseaux and Thierry Van Cutsem. Quasi steady-state simulation diag-
nosis using Newton method with optimal multiplier. In 2006 IEEE Power Engi-
neering Society General Meeting, PES. IEEE Computer Society, 2006.

[16] Konrad Purchala, Leonardo Meeus, Daniel Van Dommelen, and Ronnie Belmans.
Usefulness of DC power flow for active power flow analysis. In 2005 IEEE Power
Engineering Society General Meeting, volume 1, pages 454–459, 2005.

[17] DIgSILENT GmbH. PowerFactory User Manual. Technical report, 2020.

[18] SIEMENS. PSSE Program Operation Manual. Technical report, 2017.

[19] MATPOWER. User’s Manual, version 7.1. Technical report, 2020.

44