386
1, JANUARY 2018
A Projective Integration Method for Transient
Stability Assessment of Power Systems With
a High Penetration of Distributed Generation
Chengshan Wang, Senior Member, IEEE, Kai Yuan, Student Member, IEEE,
Peng Li, Member, IEEE, Bingqi Jiao, and Guanyu Song
Abstract—The impact of numerous distributed genera-
tors (DGs) coupled with the implementation of virtual inertia on
the transient stability of power systems has been studied exten-
sively. Time-domain simulation is the most accurate and reliable
approach to evaluate the dynamic behavior of power systems.
However, the computational efficiency is restricted by their multi-
time-scale property due to the combination of various DGs and
synchronous generators. This paper presents a novel projec-
tive integration method (PIM) for the efficient transient stability
simulation of power systems with high DG penetration. One pro-
cedure of the proposed PIM is decomposed into two stages, which
adopt mixed explicit–implicit integration methods to achieve both
efficiency and numerical stability. Moreover, the stability of the
PIM is not affected by its parameter, which is related to the step
size. Based on this property, an adaptive parameter scheme is
developed based on error estimation to fit the time constants
of the system dynamics and further increase the simulation
speed. The presented approach is several times faster than the
conventional integration methods with a similar level of accu-
racy. The proposed method is demonstrated using test systems
with DGs and virtual synchronous generators, and the perfor-
mance is verified against MATLAB/Simulink and DIgSILENT
PowerFactory.
Index Terms—Distributed generation, transient stability,
projective integration method, differential-algebraic equa-
tion (DAE), stiff, adaptive parameter.
I. I NTRODUCTION
OWER generation has been changing from a centralized,
P large power generating facility to distributed generation
involving sources of smaller capacity. The majority of these
sources require inverters on the front end when connected to
the grid. However, inverters do not have a rotating mass and
thus have low inertia. Traditionally, the voltage and frequency
Manuscript received September 27, 2015; revised January 16, 2016;
accepted March 28, 2016. Date of publication April 12, 2016; date of current
version December 21, 2017. This work was supported in part by the Project of
Integrated Operation and Planning for Smart Electric Distribution Networks,
Engineering and Physical Sciences Research Council/National Natural Science
Foundation of China Project under Grant 51261130473, in part by the
Ph.D. Programs Foundation of the Ministry of Education of China under Grant
20120032130008, and in part by the National Key Technology Research and
Development Program of China under Grant 2013BAA01B03.
The authors are with the Key Laboratory of Smart Grid of the
Ministry of Education, Tianjin University, Tianjin 300072, China
(e-mail: cswang@tju.edu.cn; yuankai83@tju.edu.cn; lip@tju.edu.cn;
bqjiao@tju.edu.cn; gysong@tju.edu.cn).
Color versions of one or more of the figures in this paper are available
online at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TSG.2016.2553359
1949-3053 c 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
IEEE TRANSACTIONS ON SMART GRID, VOL. 9, NO.
stability of power grids are mainly supported by conventional
rotating synchronous generators with large inertia coupled with
a high short-circuit current ratio (SCR). Thus, the gradual
substitution of synchronous generators by the inverter-based
distributed generators (DGs) may result in poor transient
responses of power systems during large disturbances [1] If
these issues are not well addressed, the poor responses may
develop into a transient stability problem [2], [3]. To address
this situation, the concept of virtual inertia has been developed
to improve the stability of power systems [4]–[6], wherein
the “virtual synchronous generator” (VSG) [6]–[8] is the most
popular one. The VSG concept provides new control strate-
gies for controllable DGs, such as batteries, to make them
behave similar to a synchronous generator and provide vir-
tual inertia to power grids. Thus, the integration of DGs and
VSGs can affect the dynamic behavior of the power system
as a whole.
Time-domain simulation is the most accurate and reliable
approach to evaluate the dynamic behavior of power systems.
However, the computational efficiency of such simulations is
severely limited by the multi-time-scale property of power
systems. The multiple time scales of conventional power sys-
tems mainly come from the different dynamics of synchronous
generators and their regulators, such as automatic voltage reg-
ulators (AVRs) and governors, induction motors, and static
VAR compensators (SVCs). Their time constants range from
hundreds of milliseconds to a few seconds [9] Meanwhile,
inverter-based DGs have fast dynamic responses, and the time
constants of DGs for transient stability analysis vary from hun-
dreds of microseconds to several milliseconds. Moreover, DGs
using the VSG controller involve relatively slow dynamics,
as they are controlled to supply virtual inertia. As a result,
the integration of various DGs converts the fast components
of traditional power systems into slow components, and the
time scale difference between the fastest and slowest dynamics
becomes more obvious. This time scale difference is referred
to as stiffness [10], and such stiff problems require specialized
numerical solvers to study the transient stability of power grids
in a rapid and accurate manner.
A considerable amount of effort has been taken to solve
stiff problems for transient stability simulation of conven-
tional power systems. In [11], a new integration scheme with
good numerical stability was introduced for power system
analysis; this scheme employs the exponential of the system
WANG et al.: PIM FOR TRANSIENT STABILITY ASSESSMENT OF POWER SYSTEMS
Fig. 1. Basic concept of a VSG.
matrix. This approach can obtain the precise solutions of linear
problems, and large step sizes can be applied to increase the
computational efficiency. However, using large step sizes dur-
ing fast processes may lose detailed dynamic information and
hence increase the simulation error. To solve this problem,
variable step size algorithms have been presented in [12]–[18]
so that the step size can be adaptively adjusted to the time
constants of power system dynamics. These entries in the liter-
ature often adopt implicit integration methods because of their
good numerical stability; however, such methods suffer com-
putational inefficiency with small step sizes. Then, decoupled
simulation methods were proposed in [19]–[25]. The simu-
lated system is first partitioned into a small stiff part and
a non-stiff part based on the time constants of the dynamics.
The non-stiff part is solved by explicit methods for com-
putational efficiency, whereas implicit methods are adopted
to solve the small stiff part to achieve numerical stability.
However, for a simulated system with high DG penetration,
the dimension of the stiff part grows considerably larger, and
the time savings of this decoupled method are no longer
notable. In addition, variable step size schemes are difficult to
apply to the decoupled method because the synchronization
of the step sizes between the explicit and implicit methods
should be considered. Another way to avoid the inefficiency
of variable step size algorithms is the multirate method pro-
posed in [26] and [27]: the system states are aggregated into
loosely coupled components which are then integrated indi-
vidually with a time step dictated by the time response of the
component. Unlike decoupled methods, an implicit integration
method is adopted for the integration of all the components.
Moreover, a variable partitioning strategy is presented in [28]
to modify the fast and slow partitions adaptively through-
out the simulation interval. This method is well suited for
stiff systems in which there is sustained oscillatory behavior
or systems in which only a few fast modes are frequently
perturbed.
This paper presents a projective integration method (PIM)
for the fast transient stability simulation of power systems
with high DG penetration. The main motivation is to take
the advantages of both explicit and implicit methods: effi-
ciency and numerical stability. One procedure of the PIM
consists of two stages. First, several steps of integration are
performed using explicit methods with a small step size cor-
responding to the time constants of the fast components of the
simulated system. Then, based on the previous results, a pro-
jective integration step with a larger step size is performed
using an implicit prediction-correction method. The concept
387
of the projective method was first proposed by Gear for stiff
problems with gaps in their eigenvalue spectrum [29], and
an extrapolation method, which had been proved equivalent
to the explicit Euler method, was adopted in the projective
stage. This scheme has been widely used for multi-time-scale
simulations in the chemical, thermodynamic and molecular
physical sectors. However, the numerical stability of Gear’s
projective method can be easily influenced by the step size,
and thus, it is difficult to improve the computational effi-
ciency further. The PIM proposed here adopts the implicit
prediction-correction method for the projective integration, and
the former extrapolation method is used as one component of
the predictor. Through this modification, the numerical stabil-
ity of the proposed PIM is no longer affected by its parameter
related to the step size. In addition, other properties of the
PIM and the contributions of the present paper are summarized
as follows:
1) The small-step integration in the PIM is adopted to
reflect the fast dynamics of the inverter-based DGs.
Moreover, it can also improve the accuracy of the
upcoming projective integration stage and achieve a fast
error estimation, which is applied to the adaptive param-
eter control of the PIM. Because small step sizes are
selected in this stage, explicit methods are adopted for
their computational efficiency.
2) The projective stage in the PIM is performed to reflect
the slow dynamics and overall trends of the fast dynam-
ics of the simulated system. Larger step sizes are
selected in this stage, and the implicit algorithm is used
for its numerical stability.
3) The PIM has an accuracy of order 2.
4) This paper also proposes an adaptive parameter control
scheme for the PIM based on error estimation. The error
estimation is convenient to realize through the explicit
integration steps of the PIM, and the step sizes of the
projective integration can be controlled to fit the dynam-
ics of the simulated system adaptively. This scheme can
further increase the simulation speed of the PIM without
losing accuracy.
5) We briefly introduce the transient stability model of
a VSG for the verification of the proposed PIM, which
has not been discussed before in the literature.
6) The proposed PIM is demonstrated on two test systems
with DGs and VSGs. Its computational performance is
verified using MATLAB/Simulink, a commercial tran-
sient stability simulator, i.e., PowerFactory, and various
conventional integration algorithms.
II. T RANSIENT S TABILITY M ODELING OF A VSG
The underlying idea behind the VSG technology is to mod-
ify the control strategy of an inverter-based DG so that it can
emulate the rotating behavior of a real synchronous generator
and then provide virtual inertia to power systems. The basic
concept of a VSG is shown in Fig. 1. In common practice,
energy storage devices, such as batteries, are connected to the
DC side of the VSG inverter.
388
Fig. 2. Control scheme of the VSGs for transient stability analysis.
A. Swing Equation for VSG Inertia Emulation
The main purpose of a VSG is to emulate the inertia
and damping properties of electromechanical synchronous
generators. These two main aspects can be captured by the
swing equation (1), which is widely used in the literature on
power system stability and dynamics [30].
dωg  
J = Tm − Te − D ωg − ω0
dt
where J represents the moment of inertia, D is the coefficient
of fiction loss of the synchronous generator, ωg and ω0 denote
the angular and synchronous speed of the generator, respec-
tively, Tm is the mechanical torque produced by the prime
motor, and Te is the electrical torque.
In transient stability studies, physical quantities are often
expressed in per unit (p.u.) in control loops, and power is
more controllable for power electronic devices. Therefore, the
swing equation of a VSG can be more conveniently expressed
in terms of power instead of torque, and (1) can be formulated
in p.u. by the approximation
dωg,pu  
TJ ≈ Pm,pu − Pe,pu − KD ωg,pu − 1
dt
where TJ (=2H) is the mechanical time constant, Pm,pu and
Pe,pu are the emulated mechanical input power and electri-
cal power, respectively, and KD denotes the damping constant
associated with D.
B. Control Strategy of a VSG
The model of virtual inertia must be interfaced with a power
electronic device through the VSG controller. The control
schemes of a VSG can be categorized into three main groups
based on their output references [31], namely, current ref-
erences, voltage references and power references, among
which the VSGs providing voltage references are commonly
adopted [8], [32]–[34]. In addition, the equivalence between
the swing equation (2) and the frequency-droop control cou-
pled with a first-order low-pass filter has been proven [35].
Consequently, the control strategy of the VSGs for transient
IEEE TRANSACTIONS ON SMART GRID, VOL. 9, NO. 1, JANUARY 2018
stability analysis is presented in Fig. 2, by which the active and
reactive power can be controlled separately. The active power
flow of the controller is mainly related to the inertia emulation
and the phase angle reference θ * resulting from the equivalent
swing equation, whereas the voltage amplitude reference v* is
provided by the reactive power flow. Finally, the references are
translated into PWM gate signals and supplied to the inverter.
As shown in Fig. 2, the mechanical component of a syn-
(1)
chronous generator is emulated by the VSG controller as
a logical concept, which is electrically fully effective from
a grid perspective. The damping gain KD of the emulated syn-
chronous generator in (2) is inversely linked to the droop gain
KP . Moreover, the time constant TP of the low-pass filter on
the active power flow serves as an analogous function of the
virtual inertia, which has a significant impact on the time scale
of the system. Furthermore, the parameters selected for VSG
implementations are not constrained by the physical design
of any real synchronous generator. Thus, the diversity of the
parameter selection of VSGs may aggravate the stiffness of
the transient stability problem.
Typically, the transient stability problem consists of a set
(2) of high-dimensional nonlinear differential algebraic equa-
tions (DAEs) [36]. The VSG model can be easily transformed
into these equations, as can the models of the other com-
ponents of power systems. The details of formulating these
equations can be found in [9] and [30]. Overall, the transient
stability models of power systems have the following general
form:
 dx
= f (x, y)
dt (3)
0 = g(x, y)
where the vector function f ( f : Rnx × Rny → Rnx ) relates
the components that define the dynamics of the system, the
algebraic function g(g : Rnx × Rny → Rny ) represents net-
work power flow constraints, the vector x(x ∈ Rnx ) denotes
state variables, e.g., rotor speeds and rotor angles of syn-
chronous machines and dynamic states of power electronic
devices, and the vector y( y ∈ Rny ) represents the algebraic
variables, namely bus voltage magnitudes and phases.
WANG et al.: PIM FOR TRANSIENT STABILITY ASSESSMENT OF POWER SYSTEMS
III. P ROJECTIVE I NTEGRATION M ETHOD
This section presents the projective integration method for
efficiently solving the differential equations of (3). In the
proposed PIM, one procedure consists of several explicit
steps with a small step size, followed by a projective step
with a large step size using an implicit prediction-correction
approach. We call the small step integration the inner inte-
grator, whereas the large projective step is referred to as the
outer integrator. To fairly compare the computational effi-
ciency of the PIM with the commercial simulator DIgSILENT
PowerFactory [37], the current-injection form [38], [39] of
the algebraic equations in (3) is adopted during the inner
and outer integration for the consistency of models. The alge-
braic equations are nonlinear since distribution generation is
considered during the simulation. Accordingly, we use the
Newton Raphson (NR) method [30] to solve the nonlinear
algebraic equations, which is the same as that of PowerFactory.
Specifically, one procedure of the PIM integrates from time
tn to tn+k+M as follows:
Stage I: A suitable explicit integration method is used for k
steps with a small step size t corresponding to the fast time
constants of the simulated system. Thus, the state variables
xn+k at tn+k can be computed. The explicit integration method
must have at least second-order accuracy (the reason will be
given later).
Stage II: Firstly in this stage, (3) needs to be implicitly
differenced at tn+k+M as
⎧   systems with other devices under various scenarios.
⎪ 1
⎨ xn+k+M = xn+k + T f xn+k , yn+k
2  
(4)
⎪ + f xn+k+M , yn+k+M
⎩  
g xn+k+M , yn+k+M = 0
where xn+k+M denotes the system state variables at tn+k+M ,
yn+k and yn+k+M refer to the algebraic variables at tn+k and
tn+k+M , respectively. Then, xn+k+M and yn+k+M are computed
with a large step size of T = Mt (M is a positive integer)
in the following manner:
a) The differential equations of (3) are integrated from tn+k
389
Fig. 3. Schematic diagram of one procedure of the PIM.
(1) (1)
and then compute g(xn+k+M , yn+k+M ) = 0 to correct
(0)
yn+k+M .
d) Terminate the iterations when the convergence condi-
tion shown in (8) is satisfied. If not, one can substitute
(1) (1) (0) (0)
xn+k+M and yn+k+M for xn+k+M and yn+k+M in (7)
respectively, and then repeat c) and d) until the iterations
converge.
 
 (1) (0) 
xn+k+M − xn+k+M  < ξ. (8)
In (8), ξ denotes the error threshold.
In the procedure above, the parameters k and M repre-
sent the number of inner integration steps and the multiple
of the step size t during the outer integration, respectively.
Assuming that k = 3 and M = 5, the schematic diagram of
one procedure of the PIM is shown in Fig. 3. Because the pro-
posed PIM is based on generic mathematical models, it can
also be applied to the transient stability assessment of power
IV. N UMERICAL A NALYSIS OF THE PIM
A. Numerical Accuracy of the PIM
The numerical accuracy of the PIM can be evaluated by
the local truncation error (LTE) of one procedure. Assuming
that a pth-order integration method is adopted for inner inte-
gration and that p is no less than 2 as mentioned above, the
Taylor expansion of the inner integrator from tn to tn+k can
be described as
 p    
to tn+k+M by the explicit Euler method to predict the (kt)m dm xn
initial estimates of xn+k+M as xn+k = xn + · m + O tp+1 . (9)
m! dt
  m=1
x∗n+k+M = xn+k + Tf xn+k , yn+k (5) Moreover, the Taylor expansion of the outer integrator is
and compute the algebraic function dxn+k 1 d2 xn+k
xn+k+M = xn+k + T + T 2
g(x∗n+k+M , y∗n+k+M ) = 0 to predict the initial estimates dt 2! dt2
of yn+k+M . 1 3
d xn+k  
b) Correct x∗n+k+M and compute the initial estimates of + T 3 + O T 4 (10)
4 dt3
xn+k+M through which indicates that the outer integrator is of second-order
(0) 1   accuracy. Then, the LTE of the PIM can be obtained by
xn+k+M = xn+k + T f xn+k , yn+k substituting (9) into (10) as follows:
2  

+ f x∗n+k+M , y∗n+k+M (6) L(xn+k+M , t)

 
and then the initial estimates of yn+k+M can be computed = x∗n+k+M − xn+k+M 

by solving g(x(0) (0) 
n+k+M , yn+k+M ) = 0. 1 d3 x(ξ ) 1 3
3 d xn+k
(0)
c) Further correct xn+k+M based on ≈  [(k + M)t]3 − T
 3! dt3 4 dt3
(1) 1   
xn+k+M = xn+k + T f xn+k , yn+k    p+1 
2  


(0) (0)

− O T 4 − O tm , ξ ∈ tn , tn+k+M (11)
+ f xn+k+M , yn+k+M (7) m=3 
390
Fig. 4. Stability domains for the PIM and the PIM-ref with k = 3, M = 5.
where x∗n+k+M denotes the exact solutions of the state variables
at tn+k+M . Consequently, (11) demonstrates that the PIM has
an accuracy of order 2 only if p is no less than 2, which
explains why an integration method of order no less than
2 should be selected as the inner integrator of the PIM.
B. Absolute Stability of the PIM
The explicit 4-stage Runge-Kutta (RK) method of order 4 is
implemented as the inner integrator to illustrate the absolute
stability of the PIM in this section, while the PIM with other
inner integration methods has similar properties and thus is
no longer analyzed here. The absolute stability of the PIM is
discussed when solving the Dahlquist test equation [10] as
dx(t)
= λx(t). (12)
dt
The PIM is said to be stable if the error propagation of
each procedure satisfies the absolute stability condition [10].
The error propagation is represented by the error amplification
factor as
2 + Mtλ k
σ (tλ) = ρ (tλ) (13)
2 − Mtλ
where ρ(tλ) is the error amplification factor of each inner
integration step. In this case, ρ(tλ) can be written as
(tλ)2 (tλ)3 (tλ)4
ρ(tλ) = 1 + tλ + + + . (14)
2! 3! 4!
Assuming that k = 3 and M = 5, the stability domain of the
PIM in the complex tλ-plane can be obtained by (13) and
the absolute stability condition, as depicted in Fig. 4(a). This
figure shows that the stability domain of the PIM is bounded
and nearly identical to that of the explicit 4-stage RK method
of order 4, i.e., the inner integrator of the PIM. Furthermore,
the stability domain of the PIM in the reference [29] (PIM-
ref) with k = 3 and M = 5 is shown in Fig. 4(b). These
two figures show that the stability domain of the PIM-ref
can be enhanced by the proposed modifications with the same
parameters.
Moreover, the stability domains of the PIM with k = 3 and
M ranging from 1 to 10,000 are shown in Fig. 5(a), which
barely change as M increases. Therefore, we conclude that
IEEE TRANSACTIONS ON SMART GRID, VOL. 9, NO. 1, JANUARY 2018
Fig. 5. Stability domains for the PIM and the PIM-ref with k = 3.
the absolute stability of the PIM is irrelevant to its parameter
M and that the PIM can be stable so long as the step size t
satisfies the absolute stability of its inner integration method.
Meanwhile, the stability domains of the PIM-ref with k = 3
and M ranging from 1 to 15 are shown in Fig. 5(b), which
shrink and split into two separate domains as M increases.
Consequently, it can be concluded that the numerical stability
of the PIM in [29] can be improved through the modifications
of this paper.
V. A DAPTIVE PARAMETER C ONTROL FOR THE PIM
Stability domains of the conventional integration algorithms
with A-stability, such as the trapezoidal method, are not
affected by their step sizes. Thus, some variable step size
methods can be adopted for these algorithms to accelerate
the calculation speed regardless of stability. Similarly, the sta-
bility domain of the PIM is irrelevant to its parameter M,
and accordingly, we propose an error estimation method and
an adaptive parameter control scheme for the PIM in this
section.
A. Error Estimation for the PIM
The error estimation for the PIM needs to be discussed
separately for different orders of its inner integration method.
1) Second-Order Integration Method: We assume that the
modified Euler (ME) method is selected as the inner integra-
tor of the PIM. The third-order term in the Taylor expansion
of the ME method from time nt to (n + a)t can be
described as
1 d3 xn
ηInner = (a − 1)a(a + 1)t3 3 . (15)
6 dt
Then, given parameter k, the third-order term in the Taylor
expansion of the PIM proceeding one overall integration step
with parameter M is obtained as
 
k(k − 1)(k + 1) k2 k 1
ηPIM (M) = + M + M2 + M3
6 2 2 4
3
d xn
t3 3 (16)
dt
WANG et al.: PIM FOR TRANSIENT STABILITY ASSESSMENT OF POWER SYSTEMS 391

Algorithm 1 Adaptive Parameter Control for the PIM

1: complete the inner integration and obtain xn+1 , xn+2 , and xn+3 ;
2: if l = 1
3: Ml<old> = 1;
4: else
5: Ml<old>
  =Ml−1 , end if;
 <old> 
6: T̂l = T Ml ; % a priori error estimation
⎢ ⎥
⎢  1 ⎥
⎢  3 ⎥
7: Ml = ⎣ τ  Ml<old> ⎦;
T̂l

8: proceed with stage a) to stage c) of the outer integrator and obtain

(0) (1)
xn+k+M and xn+k+M ;
 l l 
 (1) (0) 
9: T̂l = xn+k+M − xn+k+M ; % a posteriori error estimation
  l l
 
10: if T̂l  = 1.1τ
11: finish the rest of the stages of the outer integrator;
12: else
13: reject Ml , set Ml<old> = Ml , and Line 7 to Line 13 are Fig. 6. Benchmark low-voltage micro-grid network.
recomputed;
14: end if;

“floor” function, taking the integer part of x and ignoring the

decimal part, and τ in Line 10 denotes the estimated error
and thus the LTE of the PIM can be approximated by
tolerance.
neglecting the high-order residuals as
The a posteriori error estimation of the PIM-AP from Line 7
1  d3 xn   to Line 14 in Algorithm 1 is used to verify the parameter
T(M) = 2k − M 3 t3 3 + O t4
12 dt obtained by the a priori error estimation and, moreover, to
1  d 3 xn reduce the iterations of the outer integrator by improving its
≈ 2k − M 3 t3 3 . (17) initial values. T̂l in Line 9 is updated by the intermediate
12 dt
results of the outer integrator, which requires no extra com-
2) Integration Method of Order p (p > 2): In this case, (15)
putation and barely affects the simulation speed. In addition,
becomes
a little leeway for the estimated error is given in Line 10
1 d 3 xn to avoid meaningless rejections and recalculations when T̂l
ηInner = (at)3 3 (18)
3! dt swings around τ .
and further, the LTE can be estimated as (19) in a similar way.
1 3 3 d3 xn VI. C ASE S TUDIES
T(M) = − M t . (19)
12 dt3 This section presents the transient stability simulation
d3 xn results of two test systems with DGs and VSGs. The pro-
The differential term in (17) and (19) can be estimated
dt3 posed PIM-AP is implemented in a C++ simulation program
by the Lagrange interpolation formula [40] as
integrated with models of DG and VSG devices. The explicit
d 3 xn 1 4-stage RK method of order 4 is used as the inner integrator
≈ − 3 (xn − 3xn+1 + 3xn+2 − xn+3 ). (20)
dt 3 t of the PIM in this section, and our simulation experiments are
It is worth noting that the parameter k of the PIM should be conducted on a desktop computer with an Intel(R) Core(TM)
no less than 3 to ensure the solvability of (20). i5-3470 CPU running at 3.20 GHz and 4 GB of RAM.

B. Adaptive Parameter Control Strategy A. Case Study I

The main concept of the PIM with adaptive parameter con-
trol (PIM-AP) is as follows. First, after the inner integration
of the lth step, the parameter Ml of the upcoming outer inte-
grator is obtained by a priori error estimation based on Ml−1 ,
namely, the parameter of the last step of the PIM-AP. Then,
we proceed with the first three stages of the outer integrator
using Ml and estimate the local error of the lth step through
a posteriori error estimation. If the local error satisfies the
error condition, Ml is accepted and the rest of the stages will
be performed. Otherwise, Ml is rejected and the outer integra-
tor will be recalculated with an updated parameter. The pseudo
code for the adaptive parameter control of the PIM is provided
in Algorithm 1. In Line 7 of Algorithm 1, x represents the
1) Introduction of the Study Case: The 0.4 kV, 50 Hz
micro-grid prototype shown in Fig. 6 is chosen as the first
case. The main grids in both of the study cases are modeled
as a synchronous generator using the 2nd order model with
a large moment of inertia J, while the coefficient of fiction
loss D is set to zero in order to emulate the inertia property
of the upper level grids. The parameters of the micro-grid are
provided in [41], and the voltage of the main grid is set to
1.02 in per unit. This case includes two types of DG units,
namely, photovoltaic cells [42] and a battery [43] coupled
with the VSG controller. The control strategies, capacities and
power of the DG units operating at the steady-state are listed
in Table I.
392 IEEE TRANSACTIONS ON SMART GRID, VOL. 9, NO. 1, JANUARY 2018

TABLE I
C ONTROL S TRATEGIES , C APACITIES AND P OWER
OF DG S AT S TEADY-S TATE

2) Simulation Results: The simulation time is set to 5 sec-

onds, and the active power reference P∗ of the VSG unit is
changed from 0 to 0.1, in per unit, at 1.0 s under the condi-
tion that the reactive power is 0. In this scenario, the active
power step response of the VSG is featured, and the PIM-AP is
verified by the electromagnetic simulator MATLAB/Simulink
with the ode23 solver [44]. The maximum value of the param-
eter M of the PIM-AP is set to 40, and the simulation results
are shown in Fig. 7(a). The results of the PIM-AP are highly
similar to those of Simulink.
Another test is carried out to verify the applicability of the
PIM-AP to the simulation of structural transformation. The
simulation time is still set to 5.0 s, and the circuit breaker
is opened at 2.0 s, making the micro-grid transit from grid-
connection mode to islanded mode. The simulation results of
the active power and the root mean square (RMS) values of
the voltage of the VSG unit are shown in Fig. 7(b), which
shows that when the micro-grid is disconnected from the main
grid, the VSG unit adjusts its output power rapidly and the
micro-grid reaches a steady-state after a short transient time,
realizing a seamless transition. The maximum overshoot in the
transient response of the bus voltage is 3.27%. Fig. 7(b) illus-
trates that the simulation results of the PIM-AP and Simulink
are again identical and thus the PIM-AP can precisely reflect
the dynamic characteristics of each device during the mode
transition of the micro-grid.
3) Simulation Efficiency Test: The computational costs of
MATLAB/Simulink, the PIM-AP and the PIM with fixed
parameter (PIM-FP) simulating the mode transition of the
micro-grid are summarized in Table II. The Simulink-based
electromagnetic simulator requires approximately 9 min,
whereas the PIM-based transient stability simulator consumes
less than 1 s, making it more than 2,300 times faster than the
Simulink approach. The time cost of the PIM-FP decreases as
M increases or k decreases. However, when M = 9, the PIM-
FP no longer converges during the mode transition. In contrast,
the parameter M of the PIM-AP can be selected adaptively, as
shown in Fig. 7(c); large M are adopted to increase the simula-
tion speed when the micro-grid is operating at the steady-state,
and M can be rapidly reduced during the mode transition to Fig. 7. Verification of the PIM-AP implemented for testing case I.
fulfill the error conditions.
We adopt the simulation results of the explicit 4-stage RK
method of order 4 with a step size of 0.1 ms as the refer- the PIM-AP reduce rapidly to approximately 10 times smaller
ence. Then, the absolute errors of the PIM-FP and PIM-AP than those of the PIM-FP, which means that the PIM-AP can
can be obtained, as shown in Fig. 7(d) on logarithmic coor- reflect the fast transient responses of the testing system more
dinates. The simulation errors of the PIM-FP and PIM-AP precisely due to the adaptive parameter control.
remain of a similar order when the micro-grid is operating at In addition, although the quasi-state model of the inverters is
the steady-state. Then, during the mode transition, the errors of adopted in the PIM-based transient stability simulation, it can
WANG et al.: PIM FOR TRANSIENT STABILITY ASSESSMENT OF POWER SYSTEMS
TABLE II
C OMPUTATIONAL C OSTS OF D IFFERENT S IMULATORS
Fig. 8. Modified IEEE 123-node test feeder.
still precisely reflect the dynamics of the DGs and the transient
stability of the test system compared with MATLAB/Simulink
utilizing the detailed model, as shown in Fig. 7(a) and (b).
Moreover, the PIM-AP is thousands of times faster than
the Simulink-based electromagnetic simulation. Therefore, the
PIM-based transient stability assessment is suitable for stud-
ies that mainly consider the external dynamic characteristics
of power electronic devices, such as control strategy valida-
tion and stability analysis, owing to its considerable speed
advantage and acceptable accuracy.
B. Case Study II
The 4.16 kV, 50 Hz modified IEEE 123-node test feeder in
Fig. 8 is studied to test the application of the PIM-AP in the
transient stability assessment of power systems with high DG
penetration. The parameters of the testing system are shown
in [45]. A total of 85 PV units are integrated to each load
node, marked by the dotted squares as shown in Fig. 8, with
the active power outputs of half of the connected loads,., the
penetration of PV is 50%. The PV units are also modeled
as [42] and the remaining loads are supplied by the main
grid. Moreover, two VSGs with a capacity of 100 kVA are
connected to nodes 60 and 114, and the control parameters
of the VSG’s inertial emulation control loop are set differ-
ently to study their influences on the dynamic responses of the
VSG. The control parameters for the inverter control loops of
the two VSG units are listed in Table III.
The simulation time is set to 8 s, and the initial illumina-
tion intensity is 1,000 W/m2 , which changes to 1,300 W/m2
393
TABLE III
PARAMETERS FOR VSG C ONTROLLERS
Fig. 9. Simulation results of the PIM-AP and DIgSILENT PowerFactory.
at 2.0 s. Then, a three-phase fault with a grounding resistance
of 0.3  is initiated at 4.5 s and cleared after 5 cycles (0.1 s)
from the fault incidence. The PIM-AP is verified by the com-
mercial transient stability simulator DIgSILENT PowerFactory
in this case. The simulation results are shown in Fig. 9, which
indicates that the results of the PIM-AP and PowerFactory are
indistinguishable. The VSG units respond rapidly and reduce
the impact of large disturbances on the integrated system.
Additionally, Fig. 9(b) shows that the load sharing and inertial
characteristics of VSGs with different parameters are quite dif-
ferent and that parameter tuning of VSGs must be performed
according to the practical requirement.
The computational costs of different transient stability algo-
rithms are summarized in Table IV, wherein the step sizes for
these algorithms are set to 0.5 ms because of the high pen-
etration of inverter-based DGs [46], [47]. It is worth noting
that the numerical instability will occur if too large step sizes
are chosen for the explicit methods. Moreover, the adaptive
parameter control is adopted for the PIM method, and the
variable step size function is also used during the simulation
of PowerFactory. The tolerance value of the NR method is
set to 0.01% for all the case studies. As shown in Table IV,
394
TABLE IV
C OMPUTATIONAL C OSTS OF D IFFERENT
T RANSIENT S TABILITY A LGORITHMS
the simulation cost of the proposed PIM-AP is 9.53 s, which
is 4.7 times faster than the 4-stage RK method of order 4,
2.8 times faster than the ME method, 5.2 times faster than
the trapezoidal method and 2.4 times faster than PowerFactory
with adaptive step size control. The aforementioned algorithms
are all commonly used in existing commercial transient sta-
bility simulators, and our PIM-AP still performs better when
implemented for the simulation with high DG penetration.
VII. C ONCLUSION
In this paper, we have presented a projective integration
method for solving transient stability problems with multiple
time scales. The proposed PIM decomposes one procedure
into several small steps corresponding to the time constants of
the fast dynamics and one projective step for accelerating the
simulation speed, which are treated separately with different
methods. The PIM is a second-order method, and its numer-
ical stability is not affected by the selection of the step size.
Based on this property, we have also presented an adaptive
parameter control strategy for the PIM to further improve its
computational performance. Various simulation experiments
were performed on test systems with DGs and VSGs, and the
accuracy of the PIM was verified against that of electromag-
netic simulation using the MATLAB/Simulink tool. It was also
shown that the PIM is as accurate as and several times faster
than the commercial transient stability simulator DIgSILENT
PowerFactory and other conventional time-domain simulation
methods. Consequently, the proposed PIM can combine the
advantages of explicit and implicit methods to achieve both
numerical stability and efficiency. It is expected to be suitable
for the efficient transient stability analysis of power systems
with a high penetration of DGs.
R EFERENCES
[1] J. G. Slootweg and W. L. Kling, “Impacts of distributed generation
on power system transient stability,” in Proc. IEEE Power Eng. Soc.
Summer Meeting, Chicago, IL, USA, 2002, pp. 862–867.
[2] A. M. Azmy and I. Erlich, “Impact of distributed generation on the
stability of electrical power system,” in Proc. IEEE Power Eng. Soc.
Gen. Meeting, San Francisco, CA, USA, 2005, pp. 1056–1063.
[3] L. Meegahapola and D. Flynn, “Impact on transient and frequency sta-
bility for a power system at very high wind penetration,” in Proc. IEEE
Power Eng. Soc. Gen. Meeting, Minneapolis, MN, USA, 2010, pp. 1–8.
[4] N. Soni, S. Doolla, and M. C. Chandorkar, “Improvement of transient
response in microgrids using virtual inertia,” IEEE Trans. Power Del.,
vol. 28, no. 3, pp. 1830–1838, Jun. 2013.
[5] M. F. M. Arani and E. F. El-Saadany, “Implementing virtual inertia in
DFIG-based wind power generation,” IEEE Trans. Power Syst., vol. 28,
no. 2, pp. 1373–1384, May 2013.
IEEE TRANSACTIONS ON SMART GRID, VOL. 9, NO. 1, JANUARY 2018
[6] H. P. Beck and R. Hesse, “Virtual synchronous machine,” in Proc. 9th
Int. Conf. EPQU, Barcelona, Spain, 2007, pp. 1–6.
[7] J. Driesen and K. Visscher, “Virtual synchronous generators,” in Proc.
IEEE Convers. Del. Elect. Energy 21st Century Power Energy Soc. Gen.
Meeting, Pittsburgh, PA, USA, 2008, pp. 1–3.
[8] Q. C. Zhong and G. Weiss, “Synchronverters: Inverters that mimic
synchronous generators,” IEEE Trans. Ind. Electron., vol. 58, no. 4,
pp. 1259–1267, Apr. 2011.
[9] P. Kundur, Power System Stability and Control. New York, NY, USA:
McGraw-Hill, 1994.
[10] E. Hairer and G. Wanner, Solving Ordinary Differential Equations II.
Berlin, Germany: Springer-Verlag Press, 1991.
[11] J. E. van Ness and F. B. Kern, “Use of the exponential of the system
matrix to solve the transient stability problem,” IEEE Trans. Power App.
Syst., vol. PAS-89, no. 1, pp. 83–88, Jan. 1970.
[12] M. Stubbe, A. Bihain, J. Deuse, and J. C. Baader, “STAG—A new
unified software program for the study of the dynamic behaviour of
electrical power systems,” IEEE Trans. Power Syst., vol. 4, no. 1,
pp. 129–138, Feb. 1989.
[13] A. Kurita et al., “Multiple time-scale power system dynamic simulation,”
IEEE Trans. Power Syst., vol. 8, no. 1, pp. 216–223, Feb. 1993.
[14] J. Y. Astic, A. Bihain, and M. Jerosolimski, “The mixed Adams-BDF
variable step size algorithm to simulate transient and long term phe-
nomena in power systems,” IEEE Trans. Power Syst., vol. 9, no. 2,
pp. 929–935, May 1994.
[15] D. Fabozzi and T. Van Custem, “Simplified time-domain simulation of
detailed long-term dynamic models,” in Proc. IEEE Power Energy Soc.
Gen. Meeting, Calgary, AB, Canada, 2009, pp. 1–8.
[16] S. Cole and R. Belmans, “MatDyn, a new MATLAB-based toolbox for
power system dynamic simulation,” IEEE Trans. Power Syst., vol. 26,
no. 3, pp. 1129–1136, Aug. 2011.
[17] F. Chuan, “High-speed extended-term time-domain simulation for online
cascading analysis of power systemalysis of power system,” Ph.D. dis-
sertation, Dept. Elect. Comput. Eng., Iowa State Univ., Ames, IA, USA,
2011.
[18] C. Fu, J. D. McCalley, and J. Tong, “A numerical solver design
for extended-term time-domain simulation,” IEEE Trans. Power Syst.,
vol. 28, no. 4, pp. 4926–4935, Nov. 2013.
[19] M. Pöller and M. Schmieg, “Exploiting multiple time scale properties
for fast simulation algorithms,” in Proc. 13th Power Syst. Comput. Conf.,
Gomaringen, Germany, 1999, pp. 1–7.
[20] E. Hofer, “A partially implicit method for large stiff systems of
ODE’s with only few equations introducing small time-constants,”
SIAM J. Numer. Anal., vol. 13, no. 5, pp. 645–663, 1976.
[21] Y. Zhou and V. Ajjarapu, “A novel approach to trace time-domain tra-
jectories of power systems in multiple time scales,” IEEE Trans. Power
Syst., vol. 20, no. 1, pp. 149–155, Feb. 2005.
[22] D. Yang and V. Ajjarapu, “A decoupled time-domain simulation method
via invariant subspace partition for power system analysis,” IEEE Trans.
Power Syst., vol. 21, no. 1, pp. 11–18, Feb. 2006.
[23] D. Yang, “Power system dynamic security analysis via decoupled time
domain simulation and trajectory optimization,” Ph.D. dissertation, Dept.
Elect. Comput. Eng., Iowa State Univ., Ames, IA, USA, 2006.
[24] D. Yang and T. Jin, “A decoupled dynamical simulation method
via modal partition,” in Proc. IEEE Power Eng. Soc. Gen. Meeting,
Tampa, FL, USA, 2007, pp. 1–8.
[25] B. Chabaud and Q. Du, “A hybrid implicit-explicit adaptive multi-
rate numerical scheme for time-dependent equations,” J. Sci. Comput.,
vol. 51, no. 1, pp. 135–157, Apr. 2012.
[26] M. L. Crow and J. G. Chen, “The multirate method for simulation
of power system dynamics,” IEEE Trans. Power Syst., vol. 9, no. 3,
pp. 1684–1690, Aug. 1994.
[27] M. L. Crow and J. G. Chen, “The multirate simulation of FACTS devices
in power system dynamics,” IEEE Trans. Power Syst., vol. 11, no. 1,
pp. 376–382, Feb. 1996.
[28] J. Chen and M. L. Crow, “A variable partitioning strategy for the multi-
rate method in power systems,” IEEE Trans. Power Syst., vol. 23, no. 2,
pp. 259–266, May 2008.
[29] C. W. Gear and I. G. Kevrekidis, “Projective methods for stiff dif-
ferential equations: Problems with gaps in their eigenvalue spectrum,”
SIAM J. Sci. Comput., vol. 24, no. 4, pp. 1091–1106, 2003.
[30] P. W. Sauer and M. A. Pai, Power System Dynamic and Stability. Upper
Saddle River, NJ, USA: Prentice Hall, 1997.
[31] S. D’Arco and J. A. Suul, “Virtual synchronous machines—
Classification of implementations and analysis of equivalence to droop
controllers for microgrids,” in Proc. IEEE PES PowerTech, Grenoble,
France, 2013, pp. 1–7.
WANG et al.: PIM FOR TRANSIENT STABILITY ASSESSMENT OF POWER SYSTEMS 395

[32] Q. C. Zhong and G. Weiss, “Static synchronous generators for dis- Kai Yuan (S’14) received the B.S. degree in elec-
tributed generation and renewable energy,” in Proc. IEEE Power Syst. trical engineering from Tianjin University, Tianjin,
Conf. Expo., Seattle, WA, USA, 2009, pp. 1–6. China, in 2011, where he is currently pursuing the
[33] T. Shintai, Y. Miura, and T. Ise, “Oscillation damping of a distributed Ph.D. degree in electrical engineering.
generator using a virtual synchronous generator,” IEEE Trans. Power His current research interests include transient
Del., vol. 29, no. 2, pp. 668–676, Apr. 2014. stability assessment and stochastic simulation of
[34] M. Guan et al., “Synchronous generator emulation control strategy for smart grid.
voltage source converter (VSC) stations,” IEEE Trans. Power Syst.,
vol. 30, no. 6, pp. 3093–3101, Nov. 2015.
[35] S. D’Arco and J. A. Suul, “Equivalence of virtual synchronous machines
and frequency-droops for converter-based microgrids,” IEEE Trans.
Smart Grid, vol. 5, no. 1, pp. 394–395, Jan. 2014.
[36] S. M. Zali and J. V. Milanović, “Generic model of active distribution
network for large power system stability studies,” IEEE Trans. Power
Syst., vol. 28, no. 3, pp. 3126–3133, Aug. 2013.
[37] DIgSILENT PowerFactory V15.0 BETA-User Manual, DIgSILENT
GmbH, Gomaringen, Germany, Nov. 2012.
[38] F. Milano, Power System Modelling and Scripting. Berlin, Germany:
Springer, 2010, pp. 179–191.
[39] F. Milano, “On current and power injection models for angle and voltage
stability analysis of power systems,” IEEE Trans. Power Syst., vol. 31, Peng Li (M’11) received the B.S. and Ph.D. degrees
no. 3, pp. 2503–2504, May 2016. in electrical engineering from Tianjin University,
[40] M. M. Spalević, Interpolation and Approximation. New York, NY, USA: Tianjin, China, in 2004 and 2010, respectively.
Springer, 2014. He is currently an Associate Professor with the
[41] S. Papathanassiou, N. Hatziargyriou, and K. Strunz, “A benchmark School of Electrical Engineering and Automation,
low voltage microgrid network,” in Proc. CIGRE Symp. Power Syst. Tianjin University. His current research interests
Dispersed Gener., 2005, pp. 1–8. include distributed generation system and microgrid,
[42] T. Esram and P. L. Chapman, “Comparison of photovoltaic array max- smart distribution system, and transient simulation
imum power point tracking techniques,” IEEE Trans. Energy Convers., and analysis.
vol. 22, no. 2, pp. 439–449, Jun. 2007.
[43] O. Tremblay, L. A. Dessaint, and A. I. Dekkiche, “A generic battery
model for the dynamic simulation of hybrid electric vehicles,” in Proc.
IEEE Veh. Power Prop. Conf., Arlington, TX, USA, 2007, pp. 284–289.
[44] L. F. Shampine and M. W. Reichelt, “The MATLAB ODE suite,”
SIAM J. Sci. Comput., vol. 18, no. 1, pp. 1–22, 1997.
[45] W. H. Kersting, “Radial distribution test feeders,” IEEE Trans. Power
Syst., vol. 6, no. 3, pp. 975–985, Aug. 1991.
[46] S. Cole, J. Beerten, and R. Belmans, “Generalized dynamic VSC MTDC
model for power system stability studies,” IEEE Trans. Power Syst.,
vol. 25, no. 3, pp. 1655–1662, Aug. 2010.
[47] S. Liu, Z. Xu, W. Hua, G. Tang, and Y. Xue, “Electromechanical tran-
sient modeling of modular multilevel converter based multi-terminal Bingqi Jiao received the B.S. degree in electrical
HVDC systems,” IEEE Trans. Power Syst., vol. 29, no. 1, pp. 72–83, engineering from Tianjin University, Tianjin, China,
Jan. 2014. in 2011, where he is currently pursuing the Ph.D.
degree in electrical engineering.
His current research interests include optimal
planning, design, and operation of microgrid.

Chengshan Wang (M’08–SM’12) received the

Ph.D. degree in electrical engineering from Tianjin
University, Tianjin, China, in 1991.
From 1994 to 1996, he was a Senior Academic
Visitor at Cornell University, Ithaca, NY, USA. From
2001 to 2002, he was a Visiting Professor at
Carnegie Mellon University, Pittsburgh, PA. He is
Guanyu Song received the B.S. degree in electrical
currently a Professor with the School of Electrical
engineering from Tianjin University, Tianjin, China,
Engineering and Automation, Tianjin University. He
in 2012, where he is currently pursuing the Ph.D.
is the Director of the Key Laboratory of Smart Grid
degree in electrical engineering.
of the Ministry of Education, Tianjin University.
His current research interests include operating
His current research interests include distribution system analysis and plan-
optimization of smart distribution system.
ning, distributed generation system and microgrid, and power system security
analysis.
Prof. Wang is an Editor of the IEEE T RANSACTIONS ON S USTAINABLE
E NERGY. He is also an Editorial Board Member of Applied Energy, and the
Journal of Modern Power Systems and Clean Energy.