Electric Power Systems Research 77 (2007) 1514–1520

On a new approach for the simulation of transients

in power systems夽
J. Mahseredjian a,∗ , S. Dennetière b , L. Dubé c ,
B. Khodabakhchian d , L. Gérin-Lajoie e
aÉcole Polytechnique de Montréal, Canada
b Électricité de France, Clamart, France
c DEI Technology, Montréal, Canada
d TransÉnergie Technologies, Hydro-Québec, Montréal, Canada
e Hydro-Québec, Montréal, Canada

Available online 10 October 2006

Abstract
This paper presents a new simulation tool named EMTP-RV. EMTP-RV is a completely new program with a new graphical user interface and
a new computational engine. The simulation uses a new matrix formulation for computing load-flow, steady state and time-domain solutions.
Theoretical advantages are emphasized and demonstrated through practical examples. An open-architecture graphical user interface (GUI) is
developed to maximize flexibility and allow creating and maintaining complex designs.
© 2006 Elsevier B.V. All rights reserved.

Keywords: Simulation tools; Numerical methods; EMTP

1. Introduction ity to incorporate ungrounded voltage sources. This has been

corrected in [3] by using modified-nodal analysis, which incor-
Since its initial concept presented in 1969 [1] the basic EMTP porates an extra row for the voltage source equation and an extra
type simulation approach remained unchanged. It is used in var- column for the voltage source current which becomes listed in
ious commercial (DCG-EMTP Version 3, EMTP-V3 [2]) and the vector of unknowns.
non-commercial packages. The main system of symmetric equa- The assumption of admittance model existence for every
tions used in [1] is given by component is a significant limitation. An ideal transformer
model, for example, does not have admittance matrix formula-
Yv = i (1) tion. This can be partially avoided by adding losses, but can then
cause ill-conditioning problems. Other limitations are found in
This is referred to in the literature as the standard nodal analy- the representation of ideal switches. An ideal switch has pure
sis formulation. It assumes that all network components can be zero resistance when closed and becomes infinity when opened.
given an admittance matrix model. Matrix Y is the admittance This is a fundamental principle in such a model to avoid super-
matrix, v the vector of unknown voltages and i is the vector fluous natural frequencies and matrix conditioning problems.
of current sources combined with history current sources for That is why system (1) is a variable rank system where closing
the trapezoidal integration method. Since v also includes known a switch eliminates a matrix column and the corresponding row.
voltage sources, Eq. (1) is actually implemented taking only The disadvantage of such manipulations is the computational
a submatrix Yn of Y for finding n unknown voltages. One of effort, especially when the number of switches and switching
the important disadvantages of this formulation is the inabil- frequency become high.
The EMTP-V3 code was written using the legacy Fortran-77
夽 Presented at the International Conference on Power Systems Transients
language and related methods. Such codes have significant limi-
tations in automatic memory management, in code organization
(IPST’05) in Montreal, Canada on 19–23 June 2005. Paper No. IPST05-139.
∗ Corresponding author. and interfacing with external programs. Modern computing lan-
E-mail address: [email protected] (J. Mahseredjian). guages and methods provide much more powerful and efficient

0378-7796/$ – see front matter © 2006 Elsevier B.V. All rights reserved.
doi:10.1016/j.epsr.2006.08.027
 J. Mahseredjian et al. / Electric Power Systems Research 77 (2007) 1514–1520 1515

This is an augmented formulation which is keeping only the

Yn part from Eq. (1). Although submatrices are identified due to
their typical contents and for symbolic explanations given below,
Eq. (2) can be viewed as a generic Ax = b system. Matrix A is not
necessarily symmetric which provides another advantage over
Eq. (1) for some device models.
For a voltage source connected between two nodes k and m,
the source equation is given by: vk − vm = vb km . It is directly
inserted into the main system by placing a 1 and a −1 in columns
k and m, respectively, of Vr . The source voltage is vb km . The
Fig. 1. EMTP-RV main architecture.
source current condition is accounted by transposition in the
submatrix Vc . If the source is disconnected, then only its diag-
environments for software development while providing crucial onal cell in Vd is set to 1 and vb km = 0. A similar approach
advantages in code evolution and maintenance. is used for entering other models. An ideal transformer with
Another major aspect in the simulation of transients is the secondary nodes k–m, primary nodes i–j and a transformation
graphical user interface (GUI). It constitutes the first and the ratio g is modeled with the equation: vk − vm − gvi + gvj = 0.
main facet directly apparent to the user. It also represents a tech- This equation is again directly accommodated in the submatrix
nological barrier into the design and simulation of more complex Dr with the equivalent row transposition appearing in Dc . Ideal
cases. Although several GUI programming approaches are pre- switches are included straightforwardly using submatrices S.
sented in the literature [4,5], the innovation path in this field When a switch (between nodes k–m) is closed: vk − vm = 0.
is towards increased level of automation, open-architecture and When the same switch is open, its current becomes 0 by setting
advanced visualization. the corresponding diagonal cell in Sd to 1. The solved system is
This paper presents a new simulation tool named EMTP- now of fixed rank and quickly reformulated at each switch sta-
RV with a new graphical user interface EMTPWorks. It super- tus change. Adding a resistance and a fixed voltage drop (diode
sedes EMTP-V3 and presents several improvements in computa- effect) to the switch, is achieved using vk − vm − Rikm = vdc .
tional engine, graphical user interface and software development This equation goes into Sr , Sd and sb .
aspects. Although there are commercial aspects in any industrial A system similar to (2) was initially proposed in [6]. The
grade software development, this presentation is kept at the tech- system above provides further generalization and introduces new
nical contribution level. non-zero submatrix possibilities.
EMTP-RV has been written from scratch without any reverse-
engineering or code recuperation from old EMTP-V3. It consti- 2.1. Nonlinear functions
tutes a historical moment in the complex evolution of EMTP
type software. It can be proven that if nonlinear device equations are lin-
earized at each solution time-point, matrix A becomes the Jaco-
2. New formulation
bian matrix of the Newton solution. This approach provides a
The development of EMTP-RV is based on the conception true-nonlinear solution to all nonlinear devices. It eliminates the
of a computational core for the solution of its system of equa- less precise pseudo-nonlinear devices and all topological restric-
tions. The main architecture is shown in Fig. 1. The core has a tions (see [7]) encountered in the compensation method used
solution step controller (organizer) which is responsible for all in [2]. Machine models are also classified as nonlinear func-
data input and communication tasks. It schedules various task tions and solved through the same iterative solver for obtaining
calls to the simulated network devices (models and options). a simultaneous solution with all nonlinear devices and in any
These are request signals received by device codes organized topological configuration. Further details on this approach will
into encapsulated modules with device objects holding data and become available in a separate paper.
methods. Devices may respond back by participating into var-
ious requests. Participation is optional since devices may have 2.2. Control system equations
different equations and needs.
The key aspect in this architecture is the ability to easily Improvements in the solution of control system equations are
accommodate device equations into the main system of network presented in a separate paper [8]. Numerical delays that existed
equations. A new system of modified-augmented-nodal analysis in [2] between control blocks due to nonlinear feedback loops,
is proposed to eliminate various classical nodal or modified- are eliminated using a Jacobian matrix formulation.
nodal analysis limitations:
⎡ ⎤⎡ ⎤ ⎡ ⎤ 2.3. Steady-state solution
Yn Vc Dc Sc vn in
⎢V ⎥ ⎢ ⎥ ⎢ ⎥ The main purpose of a steady-state solution in EMTP is to
⎢ r Vd DVD SVS ⎥ ⎢ iV ⎥ ⎢ vb ⎥
⎢ ⎥⎢ ⎥ = ⎢ ⎥ (2) initialize the network state variables for minimizing the natu-
⎣ Dr DDV Dd SDS ⎦ ⎣ iD ⎦ ⎣ db ⎦
ral response at startup in time-domain. The formulation method
Sr SSV SSD Sd iS sb is the same as in Eq. (2), only now all variables become com-
1516 J. Mahseredjian et al. / Electric Power Systems Research 77 (2007) 1514–1520

plex and devices must provide steady-state equivalents for each

frequency in the generic harmonic steady-state solution.

2.4. Load-flow solution

The steady-state solution can provide acceptable operating

conditions (power-flow) when initialized from a load-flow solu-
Fig. 2. Automatic redrawing.
tion. The augmented matrix concept is expanded to include
load-flow constraint equations and the solved nonlinear func-
tion becomes: 4. New graphical user interface
 
A AI x The innovations in this field are in the development of an
=F=0 (3)
LLA Ld xLF open architecture and generic circuit oriented interface with
extended single-line-diagram handling capability. Signals (not
The matrix A is the linear network matrix (left-hand side matrix
nodes), devices and connectivity pins are the constitutive ele-
in Eq. (2)), AI is a connectivity matrix for accounting for load-
ments. EMTPWorks is designed for manipulating extremely
flow devices, LLA and Ld provide load-flow device constraint
large networks.
equations. The unknowns are the standard network variables
T Almost all aspects of the GUI named EMTPWorks are
x = [ vn iV iD iS ] and the load-flow device cur- scriptable and modifiable by the user. The scripting language
rents I and internal voltages E: is JavaScript with EMTPWorks extensions. Connectivity with
xLF = [ IL IPQ IPV ISL EPQ EPV ESL ]
T other languages and ActiveX usage is readily available. Data
input is based on web pages. Each device has a local web address.
Above subscripts are: L for loads, PQ for PQ control sources, This approach also allows data exchange and device addressing
PV for PV control and SL for slack bus. through the Internet. It also provides quasi-unlimited capabili-
Eq. (3) presents a new approach for multiphase load-flow ties in the development of data forms through dynamic HTML
solution. It is different from earlier presentations [9] by means of programming. The fact that device data and symbol methods
a concatenation of load-flow equations with the readily available are not hard-coded provides users with maximum flexibility for
network equations. It can inherently accommodate all EMTP creating customized devices.
device models and provide the most precise initialization. Non- EMTPWorks is an object-oriented program written in C/C++
classical load-flow devices, such as frequency-dependent trans- and based on Microsoft Foundation Class library. Each network
mission lines, become automatically included. There is no need device is an object with attributes. Attributes are available to the
to perform separate and cumbersome Jacobian derivations [9] user and used for entering data and methods. Device attributes
for network models, but only for load-flow device constraint are the customization entry point. Other objects, such as device
equations. pins and signals are given attribute specification menus. This is
an open-architecture approach.
3. Implementation Scripting is extended to drawing using EMTPWorks methods
(extensions) added to JavaScript. Drawings can be automati-
The EMTP-RV design is object-oriented. Such a design has cally updated through scripts for reflecting data changes. This
several benefits in program development and maintenance. Mod- is demonstrated in Fig. 2 for the ideal transformer unit when
ern programming concepts, such as data hiding and encapsu- updating the number of secondary windings and winding data.
lation, eliminate data corruption problems caused by device In addition to scriptable drawing, a Symbol Editor is available
models and allow establishing a clear programming approach for complete customization. The example of Fig. 3 [11] makes
by gathering all model data and methods through a single object the substation drawing appear as the actual substation using the
and module. Model code legibility is dramatically increased Symbol Editor and subcircuits. An assembly of devices can be
in comparison with old Fortran-77 coding used in EMTP-V3 placed into a subcircuit with programmable masking. Subcir-
while improving computational speed. Although the core of the cuits can contain subcircuits. This is a hierarchical approach
code is programmed using Fortran-95 (new object oriented lan- with unlimited number of levels. In EMTPWorks subcircuits
guage, see [10]) for achieving best computational speed, mixed are created automatically from user selections on the screen and
language programming is allowed specially for establishing con- pin interfaces.
nections with external functions and the graphical user interface. Hierarchical designs allow visualizing study cases at a higher
Another major benefit over EMTP-V3 is case-tailored and auto- level. In the wind park study of Fig. 4, the details of the wind
matic memory allocation. It is a key issue in the software’s ability generator control circuits, the wind generator model, the wind
to solve extremely large cases and/or usage of very small time- effect representation, the static compensator (Statcom) and the
steps with transmission line models. load are hidden in subcircuits to provide a high level visualiza-
Modern programming methods and tools can significantly tion to the actual simulation. These are multilevel subcircuits
alleviate the software development process and contribute to the with masks for allowing the user to specify internal parame-
implementation of more complex algorithms. ters. Masking allows entering extensive script codes or calls to
 J. Mahseredjian et al. / Electric Power Systems Research 77 (2007) 1514–1520 1517

Fig. 3. Example of complete redrawing with Symbol Editor.

5.2. Case 2

The design view for this test case is twice the size of the
network shown in Fig. 4. The test case is explained in details in
[12]. The Statcom is a three-level Statcom with PWM control. Its
IGBT model is based on a nonlinear diode function for increased
precision. All nonlinear functions, including the asynchronous
machines are solved simultaneously with network equations.
The simulation time-step is 50 ␮s. The mean number of itera-
tions per time-point is below 3.

Fig. 4. Wind park test case with wind behavioral model. 5.3. Case 3

external functions for calculating primitive device data at any In this example (Fig. 5) five asynchronous machines are ini-
subcircuit level. tially in steady state when a motor is started. A fault on the con-
nected network bus is initiated at 9 s and islanding occurs near
5. Demonstration examples 9.15 s. Local power is delivered by the synchronous machine
SM2. Fig. 6 presents connected asynchronous machine slips.
5.1. Case 1 Machine data is taken from [13]. Machines are automatically ini-
tialized using their slip specification. The synchronous machines
The diagram of this test case is shown in Fig. 3. The sim- and the corresponding controls are also automatically initialized.
ulation results can be found in [11]. It is a demonstration of All nonlinear devices are solved simultaneously with network
numerical stability and precision in solving extremely nonlinear equations. Transformers are modeled with a nonlinear magne-
arc model problems. Handling of specific computer require- tization branch. In EMTP-RV machines are also classified as
ments, such as memory for the frequency dependent models nonlinear devices and solved through the iterative process. All
with very small time-step usage, are also confirmed. nonlinear devices are true-nonlinear and can coexist in the same
1518 J. Mahseredjian et al. / Electric Power Systems Research 77 (2007) 1514–1520

Fig. 6. Connected asynchronous machine slips.

network without artificial insertions, such as stub lines used in

EMTP-V3. The test case assembled in Fig. 5 could not be sim-
ulated in EMTP-V3.
Despite the strongly connected network of several machines,
the mean number of iterations per time-point is 3. Only 55 s (!)
of CPU time are required to simulate an interval of 50 s with a
time-step of 200 ␮s (2 GHz computer).

5.4. Case 4

The 230 kV network case shown in Fig. 7 starts from a load-

flow solution and automatically initializes network state vari-
ables and machine controls for almost immediate establishment
of steady-state operating conditions in the time-domain simula-
tion. The network has unbalanced transmission lines and unbal-
anced loading at BUS13. Multiphase load-flow constraint data
is entered through load-flow (LF) devices and is layered over
the network simulated in time-domain. Any network can be aug-
mented with load-flow devices without any other manipulations.
All machine voltage amplitudes and angles are initially arbitrar-
ily set to positive sequence data with phase-a at zero degrees.
The load-flow algorithm is started first. It converges within six
iterations and generates load-flow solution data which automat-
ically initializes the steady-state solution of the actual network
with synchronous machines. All synchronous machines and
corresponding control circuits (AVR and Governor) are auto-
matically initialized at the time-domain simulation startup. This
is demonstrated by the waveforms for speed and power shown in
Figs. 8 and 9. The initialization of synchronous machine models
is particularly more difficult under unbalanced operating condi-
tions and requires special treatments in the steady-state solution
process.

5.5. Case 5

An exceedingly complex software engineering matter with

the graphical user interface design and the computational engine
Fig. 5. Fault and system islanding test case with machines. is the capability to show and simulate very large networks within
acceptable clock time. EMTP-RV has been tested on very large
networks. The largest network assembled so far is taken from
the Hydro-Québec system. The summary printout on primitive
devices indicates: 25 synchronous machines, 63 ideal sources
 J. Mahseredjian et al. / Electric Power Systems Research 77 (2007) 1514–1520 1519

Fig. 7. Load-flow demonstration test case.

such as SVCs. The total number of devices is above 16,000 for

approximately 7500 nodes. The entire design can be shown in
EMTPWorks using six pages and has a load time (from open
command to the complete appearance) of less than 14 s on a
2 GHz computer. It requires slightly less than 40 min per actual
real-time second of simulation with an integration time-step of
50 ␮s.
There are several advantages in the ease of maintaining such a
network: unique location of network data, exclude/include func-
tions for partial simulations and capability to perform tests on
Fig. 8. Machine angular velocity (rad/s). the actual network and/or in several locations. EMTPWorks pro-
vides scripting methods and table-browser editing functions for
manipulating data without clicking on devices one-by-one.
with impedance, 1248 transformer units, 3704 RLC branches,
610 coupled branch sections, 365 transmission lines, 1600 load
models, 1500 switches, 378 nonlinear devices (arresters, nonlin- 6. Conclusions
ear inductances,. . .) and 6838 control system devices. In several
cases primitive devices are used for building specialized devices This paper presents an overview on a new simulation tool
named EMTP-RV. It is the new generation of the DCG-EMTP
software. It constitutes a new contribution in the history of
industrial grade EMTP tools. EMTP-RV has several original
solution methods that contribute to the elimination of topolog-
ical restrictions, to improving the solution of nonlinearities, to
the flawless simulation of control systems and to the automatic
steady-state initialization methods. EMTP-RV is also based on
a new graphical user interface with innovations in its open-
architecture approach and scripting methods.
The recoding from scratch of major scientific software such as
EMTP, constituted an enormous challenge. The stressed points
were research and development on improved solution meth-
ods and software engineering approaches. Modern programming
methods and tools can significantly alleviate the software devel-
opment process and result into more powerful and user-friendly
Fig. 9. Machine power (MW). programs even within a highly scientific context.
1520 J. Mahseredjian et al. / Electric Power Systems Research 77 (2007) 1514–1520

References Proceedings of the IPST 2005 (International Conference on Power System

Transients), Montréal, June 19–23, 2005.
[1] H.W. Dommel, Digital computer solution of electromagnetic transients in [13] G.J. Rogers, D. Shirmohammadi, Induction machine modeling for electro-
single- and multiphase networks, IEEE Trans. Power Apparatus Syst. 88 magnetic transient program, IEEE Trans. Energy Conv. EC-2 (4) (1987)
(4) (1969). 622–628.
[2] EMTP Development Coordination Group, Electromagnetic Transients Pro- J. Mahseredjian (M’87) graduated from École Polytechnique de Montréal
gram Version 3. with M.A.Sc. (1985) and Ph.D. (1991). From 1987 to 2004 he worked at
[3] C.W. Ho, A.E. Ruehli, P.A. Brennan, The modified nodal approach to net- IREQ (Hydro-Québec) on research and development activities related to the
work analysis, in: Proceedings of the International Symposium on Circuits simulation and analysis of electromagnetic transients. In December 2004
and Systems, San Francisco, April, 1974, pp. 505–509. he joined the faculty of electrical engineering at École Polytechnique de
[4] H.K. Hoidalen, L. Prickler, J.L. Hall, ATPDraw-graphical preprocessor to Montréal.
ATP. Windows version, in: Proceedings of the International Conference on
Power Systems Transients, IPST, Hungary, June 20–24, 1999, pp. 7–12. S. Dennetière (M’04) graduated from École Supérieure d’Electricité (Supélec)
[5] O. Nayak, G. Irwin, A. Neufeld, GUI enhances electromagnetic transients in France in 2002. He received the M.A.Sc. degree from École Polytechnique
simulation tools, IEEE Comput. Appl. Power 8 (1) (1995) 17–22. de Montréal in 2003. From 2002 to 2004 he worked at IREQ (Hydro-Québec)
[6] J. Mahseredjian, F. Alvarado, Creating an electromagnetic transients pro- on research and development activities related to the simulation and analysis
gram in MATLAB: MatEMTP, IEEE Trans. Power Del. 12 (1) (1997) of electromagnetic transients. In October 2004 he joined the research center of
380–388. EDF where his interests are currently in the field of insulation coordination and
[7] J. Mahseredjian, S. Lefebvre, X.-D. Do, A new method for time-domain power system simulations.
modeling of nonlinear circuits in large linear networks, in: Proceedings of
the 11th Power Systems Computation Conference (PSCC), vol. 2, 1993, L. Dubé (M’77) obtained the B.A. (1967) and B.E. (1972) from Sherbrooke
pp. 915–922. University and M.E. (1973) from École Polytechnique de Montréal. He devel-
[8] J. Mahseredjian, L. Dubé, M. Zou, S. Dennetière, G. Joos, Elimination oped the TACS program of EMTP and the MODELS simulation language and
of numerical delays in the solution of control systems in EMTP, in: Pro- solver. He recently worked with IREQ (Hydro-Québec) on the development of
ceedings of the IPST 2005 (International Conference on Power Systems user-defined modeling for EMTP-RV. His consulting activities are in the field
Transients), Montréal, June 19–23, 2005. of user-defined modeling methods and languages.
[9] J.J. Allemong, R.J. Bennon, P.W. Selent, Multiphase power flow solutions
using EMTP and Newton’s method, IEEE Trans. Power Syst. 8 (4) (1993) B. Khodabakhchian holds B.S. and M.S. (École Polytechnique de Montréal)
1455–1462. degrees in electrical engineering. He has over 23 years of experiences in power
[10] J. Mahseredjian, B. Bressac, A. Xémard, P.J. Lagacé, L. Gérin-Lajoie, A system studies at Hydro-Québec. He is currently in charge of the marketing and
Fortran-95 implementation of EMTP algorithms, in: Proceedings of the sales of EMTP-RV, the new Restructured Version of the DCG-EMTP.
IPST’01 (International Conference on Power Systems Transients), Brazil,
June, 2001, pp. 686–691. L. Gérin-Lajoie received his B.Sc.A in electrical engineering from École Poly-
[11] B. Khodabakhchian, J. Mahseredjian, M.-R. Sehati, M. Mir-Hosseini, technique de Montréal in 1982. From 1982 to 1985, he worked for Hydro-Québec
Potential risk of failures in switching EHV shunt reactors in some one- in System Planning. In 1985 he joined the Control and Protection team in the Sys-
and-a-half breaker scheme substations, in: Proceeding of the International tems Studies Department. His responsibilities include protection specification,
Conference on Power Systems Transients, IPST 2003 in New Orleans, modal analysis, optimization and control coordination, as well as electromag-
September 28–October 2, 2003. netic phenomena and analytic studies related to the performance of SVC and
[12] C. Abbey, B. Khodabakhchian, F. Zhou, S. Dennetière, J. Mahseredjian, G. synchronous machines on the bulk transmission system. He also contributed for
Joos, Transient modeling and comparison of wind generator topologies, in: Hydro-Québec, the validation of the new EMTP-RV software.