Power flow algorithms for three-phase unbalanced distribution systems

Sandoval Carneiro, Jr. a, Débora Rosana Ribeiro Penidob, and Leandro Ramos de Araujob, a Federal University of Rio de Janeiro,
Rio de Janeiro, RJ, Brazil; and b Department of Electrical Energy, Federal University of Juiz de Fora, Juiz de Fora, MG, Brazil
© 2023 Elsevier Inc. All rights reserved.

Introduction 549
Why “unbalanced”? 549
Three-phase steady-state analysis 549
Previous knowledge/background 550
Chapter structure 550
Basic concepts 550
State variables 550
Descriptive system equations 550
Types of equations 550
Frame 550
Conductor representation 551
Basic classes of power flow solution algorithms 551
Approximate methods 552
Analytical methods 552
Impedance or compensation methods 552
Sweeping methods 553
Alternative methods 553
Other issues relevant to unbalanced power flow methods 554
R/X ratio 554
Robustness and precision 554
Computational time issues 555
Linear system solution 555
Sign or equation errors 555
Control equations 556
Comparison tests 556
Component models 556
Capacitors, resistors, and inductors 556
Lines and feeders 557
Loads 557
Transformers 558
Voltage regulators or line drop compensators 558
Neutral cables and groundings 558
Didactic example 558
Conclusion 560
Acknowledgment 560
References 560

Abstract

The basic foundations of three-phase and multiphase power flow for unbalanced systems and some of their algorithms are
presented in this article, which can be used in the analysis in the permanent regime (steady-state) of electrical power systems,
with special application for distribution systems. Basic concepts for understanding unbalanced power flow methods are
described. A review of the main unbalanced power flow methods is presented, considering approximate, analytical,
impedance, sweeping, or alternative methods. General algorithms for some types of methods are presented.
Particular characteristics of the methods are also discussed, issues related to their equations and implementations, their
main differences, applications, and limitations.
Some models of equipment are presented for use in multiphase unbalanced power flow. An example case is presented to
show the application of multiphase power flow in the solution of an electrical distribution system.

548 Encyclopedia of Electrical and Electronic Power Engineering, Volume 2 https://doi.org/10.1016/B978-0-12-821204-2.00078-7

 Power flow algorithms for three-phase unbalanced distribution systems 549

Introduction

Power flow studies are the determination of the voltage, current, real and reactive powers at various points in an electrical network
for a given set of steady-state operational conditions (Kersting, 2012). It is a basic tool for electric utility engineers to assist in their
tasks to perform short- and long-term operation planning and system planning.
In the early years of electrical power generation and transmission, ac calculating boards were developed as small-scale single-
phase equivalents. These ac boards were built using actual circuit elements and voltage sources. Once the set-up was complete,
measurements were taken at the points of interest and scaled back to the actual quantities thus providing the “solution”.
The rate of increase of energy production and transmission to loads of various kinds such as domestic and industrial, led to the
interconnection of formerly isolated systems. Eventually the ac boards evolved to network analyzers and the task to set up, make the
adjustments and data reading became rather formidable.
With the advent of digital computers, the first programs were written to solve for the power flow problem using positive-
sequence equivalent circuits (Tinney and Walker, 1967; Stott and Alsac, 1974). Using only the positive sequence was still necessary
to reduce memory and processing usage and increase computational performance. Algorithms for power flow solution developed
over the years are described in another article in this Encyclopedia and in classical textbooks (Stagg and El-Abiad, 1968). The
emphasis here was on balanced High and Extra-High Voltage interconnected transmission systems. Thus, the programs were not
appropriate for studies of unbalanced distribution systems. During this period, distribution systems engineers had to rely on
voltage-drop calculations using simplified models, tables and abacus.
As from the 90s the evolution of digital computers in terms of speed and storage capacities allowed for proposals of algorithms
that could solve for unbalanced DS using 3-phase equivalent models (Kersting, 2012; Cheng and Shirmohammadi, 1995; Garcia
et al., 2000). Further progress in digital technology opened an avenue of opportunities for the development of complex programs
that could cope with large scale DS, using multiphase equivalent circuits including neutrals and groundings, improved circuit
component models, control devices, and so on (Dugan and McDermott, 2011; Penido et al., 2013).
So, the algorithms had been initially applied to balanced transmission systems were expanded to solve for more challenging
configurations as will be seen in this article. Nowadays many different power flow algorithms can be applied to solve unbalanced
systems, with many different characteristics (Schneider et al., 2018).

Why “unbalanced”?
Strictly speaking, nearly all multiphase electrical networks are unbalanced, due to the simple reason that the electromagnetic
coupling between the overhead conductors depend on their relative positions.
In transmission systems the conductors are subject to rotations in which case the line is referred to as transposed. In this case the
lines can be considered balanced. In distribution systems there are normally differences among the line phases positions, so the
feeders are unbalanced.
Also, a great part of the loads in the electrical systems, especially in distribution systems, is unbalanced. The presence of three-
phase, bi-phase, and one-phase loads at the same feeder is very common. Different loads with different phase connections increase
the unbalancing of the distribution electrical systems.

Three-phase steady-state analysis

Steady-state power system analysis is mostly conducted using phasor quantities in the frequency domain, where the state variables
such as currents and voltages are represented by phasors and the circuit parameters are impedances and admittances. A phasor is
represented by a complex number that can be defined either in rectangular coordinates or in polar notation. In the former case
the phasor is defined by a real and an imaginary component, whereas in the latter case by the magnitude and angle.
The Symmetrical Components can also be used in some situations to analyze power systems in steady-state. Some simplifica-
tions may be accepted in some cases.
Positive sequence equivalent circuits are by definition symmetrical which means that the phasor quantities representing the
phase variables have identical magnitudes for the three phases and equal phase shifts between phases. This property led to the
idea to perform calculations for one of the phases only, using a single-phase to neutral equivalent circuit. Therefore, the majority
of power flow algorithms is based on this concept which applies as long as the inherent asymmetries in the system can be neglected
or made smaller, for example by transpositions of the conductors.
Most distribution systems are by nature asymmetrical, so the accuracy of algorithms based on single-phase equivalents is often
not satisfactory when applied to distribution systems. And it is common to find different feeder configurations, including neutral
cables and groundings, and different loads and equipment connections, therefore in these situations it is necessary to consider
a detailed representation of the true unbalanced nature of such systems. In these cases, a three-phase or even a multiphase repre-
sentation needs to be considered, thus a three-phase or multiphase power flow can be applied to calculate the voltages at every node
of the system and the currents in all phases, groundings, and neutral conductors.
550 Power flow algorithms for three-phase unbalanced distribution systems

Previous knowledge/background
The authors of this article have assumed that readers will be familiar with basic circuit analysis laws and methodologies in the
frequency domain (steady-state) such as Kirchhof’s Voltage and Current Law (KVL and KCL), Ohm’s Law, electromagnetic
couplings, Thevenin and Norton equivalents, nNodal network representation, per-unit system (p.u.). Basic knowledge of complex
variables and phasor representation of time-varying state variables are also expected to better understand the following sections.

Chapter structure
The objective of this article is to provide an overview of the main algorithms that have been proposed to solve for power flows in
unbalanced distribution systems where complete representations of the feeders and loads are required. In many cases it is also
necessary to represent neutral cables and grounding to achieve more precise results. It is important to say that the main applications
are in distribution systems, but they can also be applied to Transmission Systems analyses.
Initially basic concepts and a review of the main methods that have been applied to solve for the set of algebraic equations
describing the Unbalanced Power Flow problem will be presented. General algorithms will be described, and different character-
istics will be discussed. For space reasons, it is not possible to include detailed descriptions of all methods, therefore one approach
was selected to provide a vision of the steps needed to derive solution algorithms and functionalities to the Unbalanced Multiphase
Power Flow problem.
Some basic component models and equations applicable for multiphase systems will be presented. A test system will be
provided to clarify the application.

Basic concepts

This section describes some basic concepts for understanding power flow methods.

State variables
The set of variables to determine the state of the electrical system. When nodal equations are used, the main state variables are the
phase-to-ground nodal voltages for multiphase systems. As the nodal voltage is a complex value, two types of representation are
constantly used: nodal voltages in polar coordinates (mostly used in transmission systems analyses) or in rectangular coordinates
(mostly used in distribution systems analyses). Some control variables are also considered to be state variables, for example, a trans-
former tap.

Descriptive system equations

Usually descriptive equations are written using nodal equations. There are some methods that use loop equations, but these are
little used in power flow solution methods, due to the difficulty of identifying the loops and the greater complexity of the equations.
The equations are formed by complex equations, often functions of complex state variables (such as nodal voltages) and are often
written separating the real and imaginary parts. A necessary condition to unsure that the resulting system of equations is possible
and determined: the number of equations must be equal to the number of unknowns.

Types of equations
When nodal equations are used, the sum of currents or power in all nodes must be zero (Kirchoff’s current law). Usually and histor-
ically, for power flow studies, power injection equations are mostly used in modeling transmission systems whereas equations
based on current injections have been widely used in distribution systems analysis. If loop-based descriptive equations are used,
the sum of voltages in a loop must be zero (Kirchoff’s voltage law).

Frame
Power flow equations are generally written in abc phase coordinates or 012 sequence coordinates (symmetrical components). The
DQ0 coordinate is not widely used for power flow studies. Sequence components were widely used in methods developed for trans-
mission systems, historically only the positive sequence was adopted for trasmission systems due to the small unbalances. For
unbalanced three-phase systems, it is interesting to use phase coordinates, since the zero, positive, and negative sequences remain
coupled when the transformation is applied. In addition, it is considered more difficult to model electrical components in sequence
coordinates since the modeling in phase coordinates is closer to what exists in the physical systems.
 Power flow algorithms for three-phase unbalanced distribution systems 551

Conductor representation
Power systems are complex, having multiple conductors, such as phase wires, neutral wires, overhead grounding wires, and so on.
The representation of these extra wires (neutral, grounding) or of their effects in an explicit or implicit way in the power flow
methods is optional. It is however obvious that the greater the detail, the better the results obtained in the solution of the systems.
The main types of representation are:

• One-wire: this is generally used to model balanced systems, where only the positive sequence is used. This representation is
widely adopted in high voltage transmission systems power flow studies.
• Three-phase three-wire: this is used to model unbalanced distribution systems with explicit representation of phases a, b, and c in
three-phase methods. Other conductors can have their effects represented through additional methods, for example, the Kron
reduction, but depending on the behavior or connections of the equipment, this reduction can cause small errors in the results
and becomes an approximation.
• Three-phase four-wire: this is used to model unbalanced distribution systems with explicit representations of phases a, b, c,
neutral wire, and groundings.
• Multiphase-wire: this is truly a multiphase model and can virtually represent any type of electrical system, single-phase, two-
phase, three-phase, hexaphase, etc. It is also possible to represent neutral wires, groundings, lightning arresters, telecommu-
nication wires, the interaction between two or more voltage levels explicitly, among other modeling possibilities. A literature
review will show that just a few power flow methods have been made to achieve this. OpenDSS (Dugan and McDermott, 2011)
and the N-conductor Current Injection Method (NCIM) (Penido et al., 2013) are examples of multiphase power system
methods.
Controls: In electrical power systems, there are control actions that need to be represented in order to obtain the correct solution of
the power flow, as they interfere with the state of the electrical system. Control equations are commonly considered in two different
ways in power flow algorithms: the first consists in writing the control equation mathematically and adding it to the set of equations
to be solved; the second way is to leave the controls fixed during some steps while solving the problem (or an iteration of the power
flow), and then introduce the necessary adjustments based on the points of operation, and perform other iterations until
convergence.
Meshes: Currently, most distribution systems are constructed in a radial topology, therefore they do not have meshes, that is,
starting from any node it is not possible to go through a closed-loop and return to the same node. However, there are distribution
systems that have meshes, some in small quantities and others in larger quantities. Some classes of power flow methods have been
proposed to work only with radial systems to increase computational performance, but present poor performances (or do not work)
when applied to meshed systems.

Basic classes of power flow solution algorithms

This section presents a review of the main types, or classes, of power flow methods for unbalanced distribution systems, considering
approximate, analytical, impedance, sweep, or alternative methods. General algorithms for some classes are presented. Bearing in
mind that the power flow basically aims to determine the state variables of an electrical system, usually the nodal voltages, from
which it is possible to calculate various other quantities such as currents in all phases, electrical losses, and the power flows in
the electrical system.
Several methods, with several specific characteristics, are proposed each year, but usually the vast majority of methods follow the
basic steps indicated in Fig. 1.
In despite of the large number of methods available to solve the power flow problem, it can be considered that there are five basic
classes of methods. Each class can have different formulations and derivations. Each variation adopted within a given class of the
power flow methods (for example models, equations, solvers, etc.) directly impacts the characteristics of efficiency, convergence,
computational speed, and precision in the results. A possible classification would be: (i) approximate methods, (ii) analytical
methods, (iii) impedance or compensation methods, (iv) sweeping methods, and (v) alternative methods. The main characteristics
of these classes will be described in the next sections.

PF.1 PF.2 PF.3 PF.4

Definition of equations
Choice of system Choice of Solution of the
for the electrical
characteristics for component equations to find
network and controls
representation models to be used the state variables
representation
Fig. 1 Basic steps of power flow algorithms.
552 Power flow algorithms for three-phase unbalanced distribution systems

Approximate methods
Detailed modeling of a distribution system or its components is not a trivial process. The quality of the modeling can define the
quality of the solution. The more detailed the models, the greater the accuracy of the results. However, in some situations it is
possible and even of interest to make simplifications.
Approximate methods are those for which, on the premise, simplified component models are adopted, usually aiming to achieve
faster computational processing. They can be used to obtain approximate results, which are acceptable and sufficient for some
simpler studies.
Several simplifications may be performed in approximate power flow methods (Kersting, 2012). The main one being that three-
phase systems are considered to be fully balanced and with this, they can be modeled as a single line-to-neutral equivalent circuit or
positive sequence equivalent. Another widely used simplification consists in considering all loads as constant impedances and dis-
regarding controls, thus, it is possible to write the descriptive equations as a linear system, so the power flow problem can be solved
as presented in the books of linear electrical circuits.
Currently, these methods have been used less and less, due to the possibility and need for better modeling for more accurate
studies. In-depth discussions and details of these methods are found in Kersting (2012).

Analytical methods
Analytical methods are those where the modeling of the electric power system is done through descriptive equations, generally non-
linear, where the unknowns are the state variables. These equations are usually solved by established mathematical methods, such
as the Newton-Raphson method, the secant method, among others. The Newton-Raphson method is widely used in this class of
solution methods.
The analytical methods may have several variations depending on which equations and the state variables are chosen, or how the
controls are represented. The main variations used in the analytical methods are :

• The electrical network equations are written as power injections in polar coordinates and the state variables are the modules and
angles of the nodal voltages and control variables. This is usually applied to transmission systems.
• The electrical network equations are written as current injections in rectangular coordinates and the state variables are the real
and imaginary parts of the nodal voltages and control variables. This is usually applied to distribution systems.
Although many other types of variations have been published in the literature the efficiency must be analyzed for each new
application.
In Fig. 2, a basic flowchart used in analytical methods to solve for the multiphase power flow problem is presented (relative to
step PF.4 of Fig. 1).
Given that non-linearities are present in several equations, iterative methods of solution are usually adopted. Fig. 3 shows a flow-
chart of an analytical multiphase power flow method that uses the Newton-Raphson method. This method is called N-conductor
Current Injection Method (NCIM) (Penido et al., 2013). In NCIM, the distribution system is represented by a set of current injection
equations and additional control equations. The current injection equations are written using phase coordinates and the state vari-
ables are considered in rectangular form. The full Newton method is applied using sparse techniques to solve this set of nonlinear
current injection equations. Any control strategy can be implemented by writing the control equations and defining new state vari-
ables. In this method, the state variables vector (z) is composed with VRe (real part of nodal voltages), VIm (imaginary part of nodal
voltages), and other control variable; f(z) is called independent vector and is composed with the current injection equations and the
control functions; and J(z) is the Jacobian matrix. No extra procedure is required to find the steady-state solution of the power
system. A detailed description of NCIM is available in Penido et al. (2013).

Impedance or compensation methods

The impedance or compensation methods try to avoid the use of matrix inversion or matrix factoring in the solution process, in
order to increase the computational performance. These methods propose to represent, whenever possible, only the impedances
and linear elements of the distribution system in the impedance matrix (Z) of the electrical system (generally, the methods formu-
late the impedance matrix, Z, for each node in relation to the source node using the Kirchhoff’s Current Law (Schneider et al.,
2018)). Nonlinear elements are usually represented by an equivalent circuit consisting of an impedance and a current or a voltage
source (Norton or Thevenin equivalents), the impedances are placed in a fixed matrix and the current or voltage contributions to

AM.1 AM.2 AM.3

Initialize all state solve the set of Exit from the
variables z equations process

Fig. 2 Base algorithmdpower flow analytical methods.

 Power flow algorithms for three-phase unbalanced distribution systems 553

M.1 M.7
Initialize all state Exit from the
variables z iterative process
M.2 M.3 Y M.4
Assemble the right- Convergence test N Assemble the
hand vector: f(zh) |f(z) | < ε Jacobian matrix J(z)
M.6 M.5
Calculate:
Update variables
∆z=-J(z)-1f(z)
Fig. 3 NCIM algorithm.

complete the non-linear equations are calculated, at each iteration, and added to the independent vector (Dugan and McDermott,
2011). These methods are relatively simple to implement and are found in many commercial programs.
Fig. 4 shows a basic algorithm used in impedance or compensation methods for solving the power flow problem.

Sweeping methods
The sweeping methods are simple to implement, however, they may have several limitations in systems with controls or meshes. In
these methods, the distribution system is represented by an oriented tree, where the root corresponds to the substation. In the most
common version, initially, the Backward Sweep is performed, where the load currents and feeder currents are calculated, which are
accumulated starting from the loads until the substation. Then, the process called Forward Sweep is performed, where the new
nodal voltages are calculated, considering the voltage drops from the substation to the loads. A basic flowchart is presented in Fig. 5.
The most used variation of the sweeping method is the Backward-Forward Sweep (BFS) with the possibility of representing non-
linear equipment (Cheng and Shirmohammadi, 1995). It was initially designed for radial distribution systems and without control
devices. Over time, the BFS-type methods have been improved to allow the representation of weakly meshed distribution systems
and controls. The BFS method is based on calculations called successive sweeps.

Alternative methods
In addition to the methods listed above, several authors have proposed methods that are called alternative methods for solving the
multiphase power flow problem. These are approaches that consider different modeling or solution methods, or combinations

IM.1 IM.4
Initialize all state Assemble the
variables z impedance matrix
IM.2 IM.5
Assemble the Update de current
network equations sources
IM.3 IM.6
Compute the Calculate the new
network as state variable values
impedances and
Norton equivalents
(current sources)
IM.7
IM.8
Y N
Exit from the Convergence
process test

Fig. 4 Basic algorithmdpower flow impedance/compensation methods.

554 Power flow algorithms for three-phase unbalanced distribution systems

LM.1 LM.4
Initialize all Backward
state variables Sweep
LM.2 LM.5
Layer Forward
separation Sweep
LM.3 LM.6
Calculate nodal Update
currents variables

LM.7
N
Voltage test

Y
LM.9 LM.8
Update control N
Control test
variables

LM.10 Y

Exit from the

process
Fig. 5 Base algorithmdpower flow sweeping methods.

between the existing ones. Since they are not so common, they will not be covered in this article. But to register some examples, there
are methods based on: neural networks, fuzzy logic, holomorphic embedding method, Bertini equations, Graph Trace Analysis, to
name a few.

Other issues relevant to unbalanced power flow methods

This section discusses some selected characteristics of the methods to allow an overview and comparisons between them. These
characteristics are: issues related to equations and implementations; applications and limitations; aspects associated with the
convergence in the solution process of analytical, impedance, alternative and sweep methods. The main impacts of the issues
are usually mostly felt by those who implement these methods, especially for developments for applications in unbalanced distri-
bution systems. The issues are presented in summarized topics below.

R/X ratio
Cables in distribution systems have smaller diameters than cables in transmission systems, thus, the resistance value is increased
comparatively in distribution systems and with this, the R/X ratio is also higher in the distribution systems.
Approximately, some decades ago, researchers propose to resort to apply decoupled methods in the solution of distribution
systems. These methods were originally proposed to be used in transmission systems where R/X ratio are small. In consequence,
decoupled methods usually present convergence problems when applied to distribution systems, and their use should be avoided.
In the literature, many authors erroneously indicate that convergence problems due R/X ratio are normal in unbalanced power
flow algorithms. This generalization is not correct, and it is often related to outdated implementation approaches. Almost all
methods published in the last decade do not present this problem (Araujo et al., 2018).

Robustness and precision

Power flow methods must be robust and accurate. Approximate and alternative methods are generally not robust or accurate, they
provide an approximate answer, which can be useful for some applications.
 Power flow algorithms for three-phase unbalanced distribution systems 555

The linearization methods are based on simplifications in the distribution system model, seeking to achieve a balance between
computational performance and accuracy, although they are fast and with guaranteed convergence, they often do not present great
precision, but approximate results in real distribution systems. They are mostly used in planning studies where approximate values
are acceptable.
Analytical, sweeping, and impedance methods, when applied to adequate modeling, have excellent accuracy. The analytical
methods generally have excellent robustness.

Computational time issues

Multiphase power flow methods, in addition to being robust and precise, must be fast for various types of applications, for example,
in applications in real-time operation, or associated with optimization methods where there is a need for simulations of many
power flow cases, or to examine various points of operation.
Speed is often a concern for developers because real distribution systems have many notable poles (poles which must be rep-
resented) and each notable pole must be represented in a multiphase manner, that is, in just one pole there may be several nodes to
be represented (phases a, b, c, and neutral). Also, with the advent of smart grids technologies, it is necessary for various studies to
represent both medium and low voltage feeders, thus, the representation of only one real feeder can easily require more than 10,000
non-linear equations.
Usually, analytical methods, tend to be slower than impedance-based and sweep-based methods when solving low-load systems
and without controls. However, for loaded systems with many controls, analytical methods tend to be faster. The reason for this lies
in the fact that analytical methods usually need to perform the factorization of coefficient matrices at each iteration, which is
a computationally expensive process. The other methods (e.g. impedance and BFS) seek to avoid matrix factorization. On the other
hand, analytical methods need few iterations for most operating condition for any distribution system. Sweep and impedance
methods quickly increase the number of iterations required to achieve the solution when increasing the number of controls and
loading of the system.
Generally, the convergence of analytical methods is quadratic, thus at reduction of the mismatch (accepted error in the solution
of the equations) increases the precision of the solution but the number of iterations does not increase substantially. In most other
methods, a reduction of the acceptable mismatch usually does the number of iterations increases considerably.

Linear system solution

Some methods require routines to solve sparse linear systems. A poorly formulated sparse linear systems solution routine is one of
the main causes of problems in analytical methods, especially those based on Newton-Raphson method. Several good linear system
solution packages can be used to solve this type of problem, such as MATLAB or commercial mathematical packages.
The numerical precision of the variables can also strongly affect the solution. To solve this problem, in addition to the use of
efficient linear system solution routines, double precision computational variables, such as the double in C þþ, must be used.
The problem of variable precision is one of the main issue related to ill-conditioned systems.
The use of double-precision variables and good solution routines for linear systems practically solves all these problems, with
few exceptions.

Sign or equation errors

Unfortunately, the implementation of a multiphase power flow method cannot necessarily be considered correct if the results are
identical to the results of only one or a few test cases. In a considerable number of cases, some implementation errors can still
lead to the correct result and this often leads to an erroneous evaluation of the performance of a method or computational
implementation.
For example, analytical methods for power flow are generally much more complex in terms of implementation than other
methods, such as sweeping. Generally, first-order derivatives are necessary, and these procedures generally define lengthy equations,
with many variables and mathematical terms. Even if there is an error (for example in part of the equation or in the sign of an equa-
tion or variable) in the assembly of the Jacobian matrix, the method can converge to the correct result in several cases (this is even
used for example in the so-called dishonest Newton method) but can lead to divergence in heavy load power systems, or in the
presence of controls, or a considerably increase the number of iterations and computational times.
These errors are sometimes difficult to find and can usually be identified by a large number of iterations in systems of simple
convergence (systems without controls with little variation in voltage in relation to the starting point). Analytical methods, if prop-
erly implemented, usually require a small number of iterations to achieve the solution.
This problem is less critical in a sweep or impedance method, as the program does not usually converge when there are errors in
the equations, and these errors are usually easier to find.
A suitable solution is to program the equations with considerable attention and use software such as MATLAB to check the equa-
tions and derivatives.
556 Power flow algorithms for three-phase unbalanced distribution systems

Control equations
Control equations are easily implemented in analytical methods. New equations and associated state variables need to be defined.
The implementation of controls in sweep and impedance methods tends to be more complicated, where additional techniques
are adopted to consider controls actions, such as outer/inner extra iteration loops, additional routines, and other calculation proce-
dures are usually necessary.

Comparison tests
Several errors can occur in the development of power flow solution methods. In multiphase algorithms for unbalanced systems,
these errors are often even more common. Formulation errors, implementation errors, programming errors, etc. Therefore, it is
very important to perform many tests to compare the results and look for a possible validation tool. Currently, the standard compar-
ison program for the solution of unbalanced distribution systems is OpenDSS (Dugan and McDermott, 2011). It is also very impor-
tant that the methods have scalability characteristics, that is, they must be tested in large-scale distribution systems, such as the IEEE
8500 Node Test Feeder (Distribution System Analysis Subcommittee, 2020).

Component models

Equipment models, often called components, are needed to model the characteristics of real equipment in power flow programs.
These are nearly always based on equivalent circuits. Multiphase models are required to represent the individual phases of each
equipment or device, which may be connected to the three phases, phase-to-phase phase-to-neutral, or even neutral-to-ground
as is the case of grounding transformers. Thus single-phase, two-phase, three-phase or multi-phase models may be employed in
the representation of a given electrical system.
To develop the power flow algorithm, it is necessary to initially define the models (equivalent circuits) of the system components
that will be considered, since the system of equations is derived from their connections and parameters.
The first component models for three-phase power flow programs were proposed in Chen et al. (1991), and since
then, several other publications presented improvements or different models. Even today, more sophisticated models to repre-
sent more faithfully the characteristics of the equipment in a state-steady regime have been developed. For beginners in multi-
phase power flow studies, the work of (Chen et al., 1991), which deals with modeling, is considered an especially useful
reading.
In this section, basic models of the main components of electric power systems will be briefly presented. The imbalances of the
components, their existing phases, connections, and other specific characteristics need to be correctly represented in distribution
systems methods to obtain results consistent with reality. The better and more complete the modeling, the more accurate the results
obtained with the multiphase power flow tools can be. Thus, regardless of the method used, it is desirable to pay special attention to
the component models.
Some models of components for three-phase systems will be presented, the models presented relate voltages and currents and
other quantities, defining contributions of current injections of the elements of the distribution systems. Several methods use this
type of modeling. This type of modeling defines the so-called current injection equations that are used in several algorithms for
solution of power flow in unbalanced systems. There are other possibilities for modeling.

Capacitors, resistors, and inductors

Several components of multiphase distribution systems can be modeled by one or more elements called RLC, formed by resistors
and/or inductors and/or capacitors as shown in Fig. 6. Each RLC element can be connected between two nodes (k and m), and the
descriptive equations of the current injections in this case can be according to Eqs. (1)–(2), or when connected from node k to the
ground according to Eq. (3).
Imk;rlc ¼ ykm ðVm  Vk Þ (1)

Imk;rlc ¼ ykm ðVm  Vk Þ (2)

Vk Vm
k RLCkm m
Ikm Imk
Fig. 6 RLC model.
 Power flow algorithms for three-phase unbalanced distribution systems 557

Ik;rlc ¼ yk ðVk Þ (3)

Where:
Ikm, Imk, and Ik are the complex currents in the terminals of the component. Vk and Vm are the nodal voltage in node k and m. ykm
and yk are the admittance of the component.

Lines and feeders

Lines and feeders can be represented by the coupled n-phase p-equivalent lumped-parameter circuit as shown Fig. 7. Detailed
discussions on the parameter calculations are available in Kersting (2012). Descriptive current injection equations, based on current
injections, for a multiphase feeder connected between two buses (kn e mn) can be formulated as presented in Eqs. (4) and (5).
2 31 2 3 2 31 2 3
Z Zk1;m2 / Zk1;mn V  Vm1 Z Zk12 / Zk1n V
6 k1;m1 7 6 k1 7 6 k11 7 6 k1 7
6Z Zk2;mn 7 6 7 6 Zk2n 7 6 7
6 k2;m1 Zk2;m2 7 6 Vk2  Vm2 7 6 Zk21 Zk22 7 6 Vk2 7
Ikm;lin ¼ 6 7 6 7þ6 7 6 7 (4)
6 « 1 7 6 « 7 6 « 1 7 6 « 7
4 5 4 5 4 5 4 5
Zkn;m1 Zkn;m2 Zkn;mn Vkn  Vmn Zkn1 Zkn2 Zknn Vkn

2 31 2 3 2 31 2 3
Z Zkm12 / Zkm1n Vm1  Vk1 Z Zm12 / Zm1n Vm1
6 km1 7 6 7 6 m1 7 6 7
6Z Zkm2 Zkm2n 7 6 Vm  V 7 6 Z Zm2n 7 6 7
6 km21 7 6 2 k2 7 6 m21 Zm2 7 6 Vm2 7
Imk;lin ¼ 6 7 6 7þ6 7 6 7 (5)
6 « 1 7 6 « 7 6 « 1 7 6 « 7
4 5 4 5 4 5 4 5
Zkmn1 Zkmn2 Zkmn Vmn  Vkn Zmn1 Zmn2 Zmn Vmn

Loads
Exponential or ZIP models are generally used to represent loads in power flow algorithms. A load connected between two nodes (k
and m) is represented in Fig. 8. In this model, the demanded power is updated using the terminal voltage according to Eq. (6), and
the descriptive equations for current injections are shown in Eqs. (7) and (8). In distribution systems there are single-phase, two-
phase, three-phase loads, connected in delta, in star, with connection to earth or neutral cable, that is, there are many different
combinations of loads that need to be modeled correctly. For this, it is enough to combine the current injections of each element
existing in the respective connection nodes, based on the parameters of each phase.

k1 m1
Z k 1, m1
Z k 1,m 2
k2 m2
Zk 2,m2

Z k 1, mn
Zk 2,mn

kn mn
Z kn ,mn
Z k1n Z m1n

Z k 11 Z k 22 Z knn Zmnn Zm22 Z m11

Z k 12 Z k 2n Z m2n Z m 12

Fig. 7 Coupled n-phase p-equivalent lumped-parameter circuit.

558 Power flow algorithms for three-phase unbalanced distribution systems

k
Sk
Ikm

Vkm
S0 V0

m
Fig. 8 Load model.

 a  
jVkm j jVkm j b
Skm ¼ P0 þ jQ0 (6)
jV0 j jV0 j
 
Skm
Ikm;load ¼ (7)
Vkm
 
Skm
Imk;load ¼ (8)
Vmk
Where: S is the electrical power updated by the voltage. P0 and Q0 are the values of active and reactive power measured at the
nominal voltage V0. a and b are the coefficients for the load model.

Transformers
The procedure for representing transformers in power flow methods can be similar to that done in lines. According to the config-
urations of the transformers, the nodal admittance matrix that relates the primary and secondary currents and voltages is constructed
from the equivalent circuit, and the equations of contributions of current injections to the connection nodes are assembled.

Voltage regulators or line drop compensators

Voltage control devices that are often used in distribution systems and are basically built with an autotransformer and a load tap
changer. In this case, the tap in the winding of the autotransformer becomes a state variable in the power flow solution and can be
used to control the voltage of some point in the system (between two nodes, or between a node and the earth). The tap value can be
modified at each iteration, from the power flow solution process, depending on the method used. In some methods, the nodal
admittance matrix for this component must be recalculated at each iteration due the tap operations.

Neutral cables and groundings

The presence of neutral cables and grounding is very common in distribution systems. These elements are of great importance in the
modeling of unbalanced systems, as their correct representation impacts on the solution of the systems. Neglecting them or over-
simplifying their representation can lead to considerable inaccuracies or errors in the results.
Neutral cables are usually modeled together with the distribution feeder phases, as an additional cable, with all its parameters. In
some cases, such as in the situation of solidly grounded feeders, it is possible to embed their effects in the other phases and simplify
the modeling using the Kron method. But in general, it is interesting to represent them in an explicit way, also to be able to correctly
analyze the currents circulating in these cables and the voltages in the nodes.
The grounding elements must also be modeled with the parameters as close as possible to reality. One possible modeling for
groundings is to consider them as RLC elements whose model has been presented above.

Didactic example

The IEEE Distribution Systems Analysis SubCommittee (DSASC) has sponsored proposals for Test Systems to provide a platform for
validation of methodologies developed compute distribution systems analysis. This section presents an example of a power flow
solution for an unbalanced distribution system using the NCIM method. Thus, the numerical values of the variables obtained in
 Power flow algorithms for three-phase unbalanced distribution systems 559

the iterations of the power flow will be shown, with the aim of providing a basis for comparison and development of solution
methods.
The three-phase four-conductor feeder presented in Fig. 9 is connected to an ideal 12.47 kV (line-to-line) source and is 1830 m
long (Kersting, 2008). For didactic reasons, only 5 notable points will be considered (points that will be represented in the math-
ematical model). The neutral conductor is grounded at every point through a resistance of 100 U. The source and load are wye con-
nected and grounded by a resistance of 25 U and 100 U, respectively. The feeder serves impedance grounded wye connected
unbalanced constant PQ three-phase loads: Sa ¼ 3000 kVA at 0.90 lagging power factor; Sb ¼ 3500 kVA at 0.95 lagging power
factor; Sc ¼ 2500 kVA at 0.85 lagging power factor.
In order to be able to compare the results of computational implementation, the system in Fig. 9 was modeled using the NCIM
and the results of f(z) and Dz from the iterations are presented, the compositions of f(z) and Dz are from Eqs. (9) and (10).
In this case, the 10 MVA value and the phase-neutral base voltages were used as the base power. The initializations were made
considering the pu values of 1.0¢0 , 1.0¢120 , 1.0¢120 , 0.1¢0 , for the phase-to-ground voltages of phase a, b, c, and n (neutral)
respectively. The specified phase-to-neutral voltage (note, it is not the phase-to-ground voltage) is specified at 1.0 pu (7.2 kV). The
results of phase-to-ground nodal voltages are shown in Table 1.
hh i i
f ðzÞ ¼ IIm
a a b b c c
IRe IIm IRe IIm IRe f1a f2a f1b f2b f1c f2c (9)
n

hh i i
Dz ¼ a a
VRe b b
VIm VRe c
VIm VRe c
VIm P a Qa P b Qb P c Qc (10)
n

Where: IIm e IRe are the real and imaginary parts of the sum of the currents injected into the nodes. VRe e VIm are the real and
imaginary parts of nodal voltages. P e Q are the active and reactive powers injected by the substation in each phase. f represents the
substation control functions being f1 ¼ Re(Van-Vesp), f2 ¼ Im(Van-Vesp); Van is the phase-neutral voltage and Vesp is the specified
phase-neutral voltage.
Results of the first two iterations:
First Iteration:
f(z) ¼ [1.1851e-10; 0; 5.9253e-11; 1.0263e-10; 5.9253e-11; 1.0263e-10; 1.1851e-11; 0.020736; 2.3701e-10; 0; 1.1851e-
10; 2.0526e-10; 1.1851e-10; 2.0526e-10; 2.3701e-11; 0.005184; 2.3701e-10; 0; 1.1851e-10; 2.0526e-10; 1.1851e-10;

1830 m

1A 2A 3A 4A 5A
Substation Sa
12.47kV
A
1B 2B 3B 4B 5B
Sb
B
N C 1C 2C 3C 4C 5C
Sc

1N 2N 3N 4N 5N

25Ω 100Ω 100Ω 100Ω 100Ω

V=0
Fig. 9 Simplified NEV system.

Table 1 Voltage results in kV.

Va qa Vb Qb Vc qc Vn qn

1 7.153 0.198 7.202 120.419 7.245 120.221 0.053 152.000

2 7.069 0.177 7.138 120.961 7.189 119.891 0.011 152.249
3 6.986 0.560 7.075 121.513 7.133 119.555 0.032 28.252
4 6.902 0.953 7.012 122.074 7.078 119.214 0.074 28.072
5 6.819 1.354 6.950 122.645 7.023 118.867 0.116 27.863
560 Power flow algorithms for three-phase unbalanced distribution systems

Fig. 10 NEV systemdneutral results.

2.0526e-10; 2.3701e-11; 0.005184; 2.3701e-10; 0; 1.1851e-10; 2.0526e-10; 1.1851e-10; 2.0526e-10; 2.3701e-11;

0.005184; 0.1453; 0.3; 0.20034; 0.265; 0.23698; 0.012115; 0.10866; 0.017704; 0.1; 0; 0.1; 0; 0.1; 0].
Dz ¼ [0.0062247; 0.003422; 0.0062247; 0.003422; 0.0062247; 0.003422; 0.10622; 0.003422; 0.017874;
0.0030213; 0.0097378; 0.0158; 0.0027483; 0.00032791; 0.10125; 0.00068033; 0.029526; 0.0094672; 0.013253;
0.028176; 0.01172; 0.0040804; 0.096282; 0.0020655; 0.041172; 0.015906; 0.016762; 0.040559; 0.020696;
0.0078253; 0.091296; 0.0047991; 0.052805; 0.022327; 0.020258; 0.052961; 0.029686; 0.011553; 0.086273;
0.0075038; 0.25781; 0.13035; 0.36211; 0.11208; 0.21486; 0.15913].
Second Iteration:
f(z) ¼ [0.014483; 0.028645; 0.035621; 0.0050606; 0.011989; 0.022367; 0.0091483; 0.0012171; 0; 0; 0; 0; 0; 0; 0; 0;
0; 0; 0; 0; 0; 0; 0; 0; 0; 0; 0; 0; 0; 0; 0; 0; 0.0001907; 0.00049615; 0.0023472; 0.0011518; 0.0010007; 0.0028993; 0.0031572;
0.0022436; 0; 0; 0; 0; 0; 0].
Dz ¼ [0.00023985; 1.5272e-05; 0.00023985; 1.5272e-05; 0.00023985; 1.5272e-05; 0.00023985; 1.5272e-05;
0.00026052; 3.9146e-06; 0.00028648; 8.9951e-05; 0.00032361; 5.8918e-05; 4.8077e-05; 2.8511e-06; 0.00028122;
2.3203e-05; 0.00033313; 0.00016452; 0.00040739; 0.00013321; 0.00014358; 9.7753e-06; 0.00030185; 4.2187e-
05; 0.00037972; 0.00023941; 0.00049111; 0.0002072; 0.00033557; 2.179e-05; 0.00032235; 6.046e-05;
0.00042618; 0.00031505; 0.0005747; 0.00028045; 0.00052831; 3.2373e-05; 0.029396; 0.014437; 0.025929;
0.022328; 0.0014366; 0.023633].
Final results of nodal voltages, after two iterations, are shown in Table 1. Fig. 10 shows the results for the neutral voltage, these
apparently unusual results (V curves) are normaly found in several distribution systems.

Conclusion

This article was conceived to provide an overview of algorithms that have been developed to solve for the power flow problem in
unbalanced electrical power networks. Some algorithms, equipment models, and the main characteristics of multiphase power flow
have been presented. It has also discussed some difficulties, important elements for the implementation, use of the results, and
attempted to show the readers the main challenges that have been overcome, thanks to the numerous researchers and engineers
who have dedicated much effort to advance these tools.

Acknowledgment

The authors would like to recognize, in memoriam, the contributions of Prof. José Luiz Rezende Pereira to the development of the Current Injection
Method of Power Flow Analysis. He was a research leader and educator who inspired numerous young students to pursue meaningful research
projects. He would have been a co-author of this article but unfortunately left us prematurely due to Covid-19.

References

Araujo, L.R., Penido, D.R.R., do Amaral Filho, N.A., Beneteli, T.A.P., 2018. Sensitivity analysis of convergence characteristics in power flow methods for distribution systems. Int. J.
Electr. Power Energy Syst. 97, 211–219.
Chen, T., Chen, M., Inoue, T., Kotas, P., Chebli, E.A., 1991. Three-phase cogenerator and transformer models for distribution system analysis. IEEE Trans. Power Deliv. 6 (4),
1671–1681.
Cheng, C.S., Shirmohammadi, D., 1995. A three-phase power flow method for real-time distribution system Analysis. IEEE Trans. Power Syst. 10 (2), 671–679.
 Power flow algorithms for three-phase unbalanced distribution systems 561

Distribution System Analysis Subcommittee, IEEE PES Test Feed. Accessed Sep. 30, 2020. [Online]. Available https://site.ieee.org/pes-testfeeders/resources/.
Dugan, R.C., McDermott, T.E., 2011. An open source platform for collaborating on smart grid research. In: Proc. Power and Energy Society General Meeting, 2011. IEEE, pp. 1–7.
Garcia, P.A.N., Pereira, J.L.R., Carneiro Jr., S., Da Costa, V.M., Martins, N., 2000. Three-phase power flow calculations using the current injection method. IEEE Trans. Power Syst.
15 (2), 508–514.
Kersting, W.H., 2008. A three-phase unbalanced line model with grounded neutrals through a resistance. In: 2008 IEEE Power and Energy Society General MeetingdConversion
and Delivery of Electrical Energy in the 21st Century, pp. 1–2.
Kersting, W.H., 2012. Distribution System Modeling and Analysis, third ed. CRC Press, Boca Raton, USA.
Penido, D.R.R., Araujo, L.R., Carneiro Junior, S., Pereira, J.L.R., 2013. A new tool for multiphase electrical systems analysis based on current injection method. Int. J. Electr. Power
Energy Syst. 44 (1), 410–420.
Schneider, K.P., et al., 2018. Analytic considerations and design basis for the IEEE distribution test feeders. IEEE Trans. Power Syst. 33 (3), 3181–3188.
Stagg, G.W., El-Abiad, A.H., 1968. Computer Methods in Power System Analysis”. McGraw-Hill Inc.
Stott, B., Alsac, O., 1974. Fast decoupled load flow. IEEE Trans. Power Apparatus Syst. PAS-93, 859–869.
Tinney, W., Walker, J., 1967. Direct solution of sparse network equation by optimally ordered triangular factorization. Proc. IEEE 55, 1801–1809.