You are on page 1of 8

Proceedings of the 35th Hawaii International Conference on System Sciences - 2002

Calculation of the Probability Density Function of Critical Clearing Time


in Transient Stability Analysis

Yiqiao Liang Saffet Ayasun Chika Nwankpa

ECE Department
Drexel University
Philadelphia, PA 19104

Abstract--In this paper, the critical clearing time, tcc in power fault clearing time is sometimes employed as an index as well.
system transient stability analysis is modeled as a random Even though the direct methods have drastically improved the
variable due to the randomness nature of power system load. A program speed, the huge size of actual power systems and the
linear approximation approach that mainly involves the complexity of the transient stability problem still involve
sensitivity calculation of the derivative of tcc to system load is prohibitive calculations. However, transient stability study
developed to obtain the probability density function (pdf) of tcc will be implemented in Energy Management System (EMS)
based on the pdf of system load. This approach is verified to for real time on line operation due to the advent of more
be accurate under the condition of small load deviation by advanced algorithms and faster computers in the near future.
Monte Carlo simulation method. The probability of system Like many other application programs in the EMS system [12],
being transiently stable for particular disturbances are the dynamic security Assessment (DSA) program (transient
calculated based on the pdf of tcc. stability program, TSP) will be continually executed in a
certain interval of time. Depending on the computing power of
I. INTRODUCTION the EMS system, this interval can range from tens of minutes
to an hour. The execution flow of this program can be
Power system stability assessment is a major requirement for illustrated by Figure 1 below.
safe operation of a power system. It is divided into two main
components, namely, transient stability and steady-state TSP Starts TSP Starts

stability. A power system is “transiently stable” for a particular


TSP Ends TSP Ends
steady-state operating condition and for a particular large
disturbance if, following that disturbance, it reaches an
acceptable steady-state operating condition [1]. The
conventional method that is still adopted in transient stability
studies is via time domain simulation. This method solves the
system of algebraic and differential equations describing the tk tk+ ∆t tk+1 t
power system under different faulted conditions. Time domain
simulation techniques can be used for off-line transient Figure 1. TSP execution flow in the EMS system
stability studies and can simulate the dynamics under different
time scales such as medium and long-term dynamics. The At time tk TSP is started and it gets a snapshot of system data
method suffers in its slow speed and inability to provide an at this instant of time and performs necessary calculations.
index of stability for the entire system. After ∆t , the program terminates and conclusions regarding
the system transient stability are obtained. At time tk+1 the
Direct Methods employing the energy function and the above procedure will be repeated.
Lyapunov method provide an index for evaluating the stability
of the system quantitatively. There are several variants of the As we all know that the result of transient stability analysis
direct method such as Potential Energy Boundary Surface depends on the system load. In another words, the critical
(PEBS) method [2][8]; BCU method [3] and Extended Equal clearing time tcc is a function of system load. Thus, the
Area Criterion (EEEAC) method [4] etc. The procedures of conclusions drawn based on the system load at time tk will not
determining power system transient stability in these various always hold true during tk+ ∆t to tk+1 due to the fact that
methods are basically the same. The critical clearing time tcc is system load keeps changing. More specifically, tcc should be
calculated and compared with the actual fault clearing time. If modeled as a random variable because of the randomness of
the former is greater than the latter the system is determined to the system load. Based on the arguments above, a system
be transiently stable. The difference between these two values being transiently stable under a particular contingency is also a
can be used as an index of system transient stability. A higher random event. The probability of this random event is a very
value of this index indicates a more stable system. The important bit of information to operating engineers. Under the
difference between the critical energy value and the energy at

0-7695-1435-9/02 $17.00 (c) 2002 IEEE 1


Proceedings of the 35th Hawaii International Conference on System Sciences - 2002

assumption that the pdf of tcc is known, this probability can be reducing a power system model to an OMIB system, modeling
calculated by the equation below. fluctuations in load become even more necessary.
P{system being stable}=P{tc<tcc} (1)
2.1 Swing equations
The shaded area in Figure 2 below graphically illustrates this
In power system transient stability studies, the system under
probability value. For illustrative purposes the Gaussian
study is usually divided into three subsystems according to the
distribution is used in this figure.
time when disturbance happens. They are namely,
• Pre-fault system: system before the disturbance
• Fault-on system: system during the fault
• Post-fault system: system after the fault is cleared
pdf of critical fault
clearing time tcc For a lossless and non-salient pole OMIB system, the well-
known equation that describes the dynamic behavior is given
below.
d 2δ
M = Pm − Pmax sin δ (2)
dt 2

where in this equation,


EV
Pmax =
tc X
Figure2. Graphical illustration of the probability Pm—generator power
of system being transient stable M—the moment of inertia of the generator
δ —generator power angle
Due to the fact that the critical clearing time tcc is a function of E—generator voltage behind its transient reactance
system load, the pdf of tcc can be calculated based on the pdf V—voltage magnitude at the infinite bus
of system load. The pdf of system load can be obtained from X—equivalent reactance between the generator and
the historical data of the Supervisory Control and Data the infinite bus
Acquisition (SCADA) system. Unfortunately, obtaining the This equation holds true for Pre-fault, Fault-on and Post-fault
function of tcc with respect to system load is complex and time systems except that the Pmax will have different values for these
consuming. Hence in this paper, a simple approximation three subsystems in most of the cases. We will use Pmax1, Pmax2
method, which is linear approximation, is adopted for the and Pmax3 to represent this value for Pre-fault, Fault-on and
purpose of real time operation. Post-fault systems, respectively. The well known power curve
graphics of Pre-fault, Fault-on and Post-fault systems are
The remaining parts of this paper are organized as follows. illustrated in Figure 3 below. Where in this figure, δ 0 is the
The mathematical relationship between tcc and system load is
formulated in Section II for a One Machine to Infinite Bus Pre-fault stable operating power angle, δ 3 is the Post-fault
system (OMIB). In Section III, approaches for solving the stable operating power angle and δ c is the fault-clearing
problem are described and difficulties in these approaches are angle. The values of δ 0 and δ 3 are given by the equations
addressed and overcome. Section IV gives two examples in
(3) and (4) as follows.
which the histograms of both exact and approximated tcc are
calculated. The probabilities of system being transiently stable Pe
are calculated in this section as well. Some discussions based
Pmax1
on the results of the examples are presented in Section V. Pre-Fault
Section VI concludes the paper.
Pmax3 Post-Fault
II. PROBLEM FORMULATION Area B
Pm

In EEAC method [4], it is shown that the power system for Area A
Pmax2
studying transient stability can always be reduced to a one Fault-On

machine to infinite bus (OMIB) system. This OMIB model


will be adopted and the Equal Area Criterion [7] will be
δ0 δ3 δc π−δ3 δ
revisited to establish the relationship between tcc and system
load in this section. With the inherent errors involved in Figure 3. Pre-fault, Fault-on and Post-fault power curves

0-7695-1435-9/02 $17.00 (c) 2002 IEEE 2


Proceedings of the 35th Hawaii International Conference on System Sciences - 2002

 Pm  III. APPROACHES
δ 0 = sin −1  
 (3)
 Pmax 1  From equations (3)-(8) we can see that it is very complicated
−1 
P  and time consuming to obtain the exact pdf of tcc based on the
δ 3 = sin  m  (4)
 max 3 
P pdf of system load, Pm. For the purpose of real time
application, equation (8) will be linearized around the
δ c is determined by fault clearing time and Fault-on system operating point to obtain the approximate pdf of tcc. The
dynamic equation. Monte Carlo Simulation method will be introduced and
adopted to verify the accuracy of this approximated pdf of tcc.
2.2 Critical clearing angle and time
3.1 Linear Approximation
Critical clearing angle is defined as the maximum power angle,
such that when the fault is cleared before this angle the system The linear approximation of the critical clearing time can be
is transiently stable [1]. From equation (2) and Figure 3, we presented by the equation below.
know that area A in Figure 3 is the accelerating area while area
dt cc
B is the maximum possible decelerating area. When area A is t cc = t cc0 + (Pm − Pm0 ) (9)
less than area B the system is transient stable. When the dPm P =P
m m0

clearing angle is increased to a certain level such that area A is


equal to area B the system is at a boundary condition. Under Where Pm0 is the system operating point and tcc0 is the critical
this boundary condition, the clearing angle is called the critical clearing time under this operating point. The key issue in this
clearing angle denoted as δ cc . It can be calculated by the dt cc
linear approximation method lies in the calculation of .
equation below. dPm
Applying the chain rule of derivation to equation (8), we can
 Pm (π − δ 3 − δ 0 ) − Pmax 3 cos δ 3 − Pmax 2 cos δ 0  express this derivative as follows;
δ cc = cos −1  
 Pmax 3 − Pmax 2  dt cc ∂F dδ cc ∂F dδ 0 ∂F
= + + (10)
(5) dPm ∂δ cc dPm ∂δ 0 dPm ∂Pm

The critical clearing time is the time when the power angle As shown in equation (10) above, there are five terms that
reaches the critical clearing angle during the first swing. After compose this derivative. Four of these terms can be easily
some manipulations of equation (2), this critical clearing time obtained and they are described by the equations below.
is expressed by equations (6) and (7) below [1].
dδ 0 1
δ
= (11)
(δ ,δ 0 )dδ dPm Pmax 1 cos δ 0
δ0∫
t cc = cc f (6)
dδ cc
where
=−
1
(π − δ 3 − δ 0
2 
−1 2 dPm (Pmax 3 − Pmax 2 ) sin δ cc
f (δ ,δ 0 ) =  (Pm (δ − δ 0 ) + Pmax 2 (cos δ − cos δ 0 )) (7)
 M  Pmax 2 Pm Pm 
It is observed that lim f (δ ,δ 0 ) → ∞ which indicates that + tan δ 0 + tan δ 3 − −  (12)
Pmax 1 cos δ 0 Pmax 3 cos δ 3 
δ →δ 0
Pmax 1 
equation (6) is an improper integration. Equations (3) to (7)
imply that the critical clearing time tcc is a very complicated
∂F δ 1 −(δ − δ 0 )dδ
function of system load, Pm. This function can be written as = ∫ cc
∂Pm δ 0 M 3
equation (8) below. 2 
 M (Pm (δ − δ 0 ) + Pmax 2 (cos δ − cos δ 0 ))
t cc = F (δ 0 (Pm ),δ cc (Pm ), Pm ) (8)  
(13)
In the following section we will talk about linearization of this
∂F
equation for the purpose of obtaining the pdf of tcc promptly. = f (δ cc , δ 0 ) (14)
The accuracy of this approximation will be verified by Monte ∂δ cc
Carlo Simulation method.
In order to obtain ∂F ∂δ 0 , a new formula has to be
introduced here. Because the most widely used formula,

0-7695-1435-9/02 $17.00 (c) 2002 IEEE 3


Proceedings of the 35th Hawaii International Conference on System Sciences - 2002

δ cc
∂  ∫ f (δ ,δ 0 )dδ  ∂δ 0 = − f (δ 0 ,δ 0 ) − ∫
(
δ 0 ∂f δ ,δ 0 )
dδ is not
The procedure for applying this method to transient stability
δ
 0  δ cc ∂δ study is mainly composed of three parts. The first part
0
simulates the occurrence of the uncertainties in system load. A
valid any more in this case due to the fact that
random number generator is used to generate pseudo random
lim f (δ ,δ 0 ) → ∞ . numbers representing uncertain system load values. The
δ →δ 0
second part is to calculate the critical clearing time tcc based on
every load value that is generated in the first part. The Equal
New Formula
Area Criterion is employed in the computation. More
For a given function, G ( x ) = ∫ f (t , x )dt , its derivative can be
a
specifically, equations (3)-(7) are employed to calculate the tcc
x
values. Based on the results from the second part, the third part
obtained by the equation below,
calculates the histogram of tcc. This histogram is then used to
dG (x ) a  ∂f ( z , x ) ∂f ( z , x )  approximate the pdf of tcc. Once we obtain this pdf, the
= − f (a , x ) + ∫  + dz given that
dx x  ∂z ∂x  probability of system being transiently stable can be calculated
by applying equation (1). The flow chart of the whole
a  ∂f (z , x ) + ∂f (z , x ) dz
∫x 
f(a,x) and procedure is given in Figure 4.
∂z  exist. ∂x

The derivation of this formula is presented in the Appendix. Start
Applying this formula to equations (6) and (7), we obtain
∂F Ninf=0; i=0
as follow.
∂δ 0 N=Number of Monte Carlo Trials

∂F δ P
= − f (δ cc ,δ 0 ) + ∫ cc max 2 (sin δ − sin δ 0 ) f 3
(δ ,δ 0 )dδ i=i+1
∂δ 0 δ 0 M
(15) Obtain a random load by using a random
number generator
Plugging equations (11)-(15) into equation (10), we obtain the
derivative of tcc with respect to system load Pm below.
Solve the pre-fault system operating point
dt cc 
= − f (δ cc ,δ 0 )
1
+
dPm  max 1 cos δ 0
P
Solution Exists?
1  P No
+  π − δ 3 − δ 0 + max 2 tan δ 0
(Pmax 3 − Pmax 2 ) sin δ cc 
 Pmax 1 Yes

Solve the post-fault system operating point


Pm Pm 
+ tan δ 3 − − 
Pmax 1 cos δ 0 Pmax 3 cos δ 3 

Solution Exists?
P (sin δ − sin δ 0 ) δ − δ 0  3 No
+ ∫δδ  max 2 −  f (δ ,δ 0 )dδ
cc Yes

 MPmax 1 cos δ 0
0
M 
Calculate the system critical clearing time tcc
(16)
Ninf=Ninf+1
3.2 Monte Carlo Simulation
No
Mote Carlo simulation method [10][11] applies repeated i>N?
probabilistic trials, the collection of which can be used to Yes
estimate a probability. When the number of trials increases the
accuracy of the estimated probability increases as well. Since Calculate the histogram of tcc
this method doesn’t put any restriction on a probabilistic
model. It is very easy to implement. The principal difficulty in Calculate P{system being transiently stable}
applying this method is that it is computationally intensive. As
a result, it is not appropriate for real time application. Ninf-- Number of infeasible
However, it can be used to verify the accuracy of the linear Stop cases
approximation method that is proposed in the previous
subsection. Figure 4. Flow chart of Monte Carlo simulation

0-7695-1435-9/02 $17.00 (c) 2002 IEEE 4


Proceedings of the 35th Hawaii International Conference on System Sciences - 2002

In the next section, two examples will be given. The |Error|=100*(Pexact-Papproximated)/Pexact and Papproximated is
histograms of both the exact tcc and approximated tcc will be obtained by the proposed approximation method.
shown. The probability of system being transiently stable
versus fault clearing time will be calculated and plotted. Table 3. Exact and approximated probabilities (case 2)
tc (second) Papproximated Pexact Error (%)
IV. EXAMPLES 0.20 0.8486 0.8342 1.71792
0.22 0.7644 0.7123 7.31475
A one machine to infinite bus system will be employed to test 0.24 0.6577 0.5716 15.0629
the proposed approximation method in this section. For the 0.26 0.5325 0.4345 22.5523
purpose of comparison, the accurate results are also obtained 0.28 0.4058 0.3205 26.5954
by numerically integrating equations (5) and (6). After some
0.30 0.2898 0.2240 29.3671
calculation, the parameters in system dynamic equation
0.32 0.1886 0.1531 23.1759
(equation (2)) for Pre-fault, Fault-on and Post-fault subsystems
0.34 0.1116 0.1057 5.53386
are listed in Table 1 below.
0.36 0.0595 0.0717 17.01578
Table 1. System parameters 0.38 0.0297 0.0498 40.32511
Pre-fault Fault-on Post-fault 0.40 0.0141 0.0336 58.09446
M 0.01 0.01 0.01
Pmax 2.22 0.20 1.11 .
700
System load Pm is modeled as a Gaussian distribution random
variable. Two different cases are chosen regarding to the mean
600
value and standard deviation of this random variable. In both
cases, deterministic values of the fault clearing time are used. 500

4.1 Case 1 400

In this case, the mean value of Pm is 0.5 pu while its standard


300
deviation is 4% of its mean value. The histograms of both the
accurate tcc and linearly approximated tcc are shown in Figures
200
5 and 6, respectively. The probabilities of system being
transiently stable at different levels of fault clearing time are 100
calculated by applying equation (1) and the results are plotted
in Figure 7. The numerical values of these results are listed in 0
0.2 0.22 0.24 0.26 0.28 0.3 0.32
table 2 below. Critical clearing time tcc (second)

Table 2. Exact and approximated probabilities (case 1) Figure 5. Histogram of critical clearing time tcc
tc (second) Papproximated Pexact Error (%)
700
0.20 1.0000 1.0000 0
0.21 1.0000 0.9999 0.01000 600
0.22 0.9967 0.9966 0.01003
0.23 0.9755 0.9670 0.87901 500
0.24 0.8638 0.8196 5.39287
0.25 0.6045 0.5107 18.36694 400

0.26 0.2820 0.2058 37.02623


300
0.27 0.0718 0.0525 36.76190
0.28 0.0112 0.0088 27.27272 200

100
4.2 Case 2
In this case, mean value of Pm is 0.5pu while its standard 0
0.2 0.22 0.24 0.26 0.28 0.3 0.32
deviation is 20% of its mean value. The calculations carried Critical clearing time tcc (second)
out in case 1 are repeated in this case. The results are plotted
Figure 6. Histogram of linearly approximated tcc
in Figures 8, 9 and 10. The numerical values of these results
are listed in Table 3. In tables 2 and 3,

0-7695-1435-9/02 $17.00 (c) 2002 IEEE 5


Proceedings of the 35th Hawaii International Conference on System Sciences - 2002

1 1
exact model exact model
0.9 linear approximation 0.9 linear approximation

0.8 0.8

0.7 0.7
Probability of stability

Probability of stability
0.6 0.6

0.5 0.5

0.4 0.4

0.3 0.3

0.2 0.2

0.1 0.1

0 0
0.2 0.25 0.3 0.35 0.4 0.45 0.5 0.2 0.25 0.3 0.35 0.4 0.45 0.5 0.55 0.6
Fault clearing time tc (second) Fault clearing time tc (second)

Figure 7. Probability of system being transiently stable versus Figure 10. Probability of system being transiently
fault clearing time stable versus fault clearing time

900

800 V. DISCUSSIONS
700
Some observations from the results obtained in the previous
600 section are listed and discussed in this section.
500
First, in case of small load deviation that corresponds to case 1
400 in Section IV, the results obtained by the proposed linear
300
approximation method are accurate. Comparing Figure 5 with
Figure 6, we can see that the histograms of accurate tcc and
200
linearly approximated tcc are very close. As a result, the
100 probabilities of system being transiently stable obtained by
both the linear approximation method and the original model
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 are very close as well. Figure 7 tells this fact. The reason of
Critical clearing time tcc (second) this is when load deviation is small; the error between the
Figure 8. Histogram of critical clearing time tcc accurate tcc and the approximated tcc is also small.
Second, in case of large load deviation, which corresponds to
900 case 2 in Section IV, the error gets large. The histograms of
the accurate tcc and linearly approximated tcc are not close any
800
more. This is clear when comparing Figure 8 with Figure 9.
700 The probabilities of system being transiently stable obtained
by the linear approximation method departs from the exact
600
ones. This can be told by examining Figure 10. The above
500 results are due to the reason that when load deviation gets
400
larger the linear approximation deteriorates. There are two
ways of increasing the accuracy of the proposed approximation
300 method. One of them is to use second order approximation
200
instead of linear approximation. The other one is to decrease
the interval between every run of the transient stability analysis
100 program so that the system load deviation is decreased as well.
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 Simulation times for exact and linear approximation models
Critical clearing time tcc (second) are summarized in Table 4 for both case 1 and case 2. As can
Figure 9. Histogram of linearly approximated tcc be seen from this table that simulation time decreases
significantly when we implement the proposed linearly

0-7695-1435-9/02 $17.00 (c) 2002 IEEE 6


Proceedings of the 35th Hawaii International Conference on System Sciences - 2002

approximated model. It is clear in Table 3 that simulation


times of linearly approximated model is 83 times less than
those of the exact model for both cases, which is a very REFERENCES
significant advantage of the proposed approximation for real
time on line applications. [1] M.Pavella and P.G. Murthy, “Transient Stability of Power Systems:
Theory and Practice”, John Wiley & Sons, 1994
When this proposed approximation method is implemented in [2] Kakimoto N., Y. Ohsawa and M. Hayashi “Transient stability
the future EMS system, the procedure of system transient analysis of electric power system via Lure type Lyapunov function -
Parts I and II. Trans. IEE of Japan 98, pp. 63-79, 1978
stability analysis will be as follows. [3] Hsiao-Dong Chiang, Felix F. Wu and Pravin P. Varaiya, “A BCU
1) When the time of running transient stability analysis Method for Direct Analysis of Power System Transient Stability”
program comes, get a snapshot of system load data and IEEE Transactions on Power Systems, Vol. 9, No. 3, pp. 1194-1208,
other system information. August. 1994
[4] Xue Y., Th. Van Custsem and M. Ribbens-Pavella, “Extended equal
dt cc area criterion. Justifications, generalizations, applications” IEEE
2) Compute tcc and . Based on these two values and
dPm Trans. PWRS-4, pp. 44-52, 1989
[5] Billinton, R. and Kuruganty, P.R.S., “A Probabilistic Index for
equation (9), formulate the approximation model and Transient Stability”, IEEE Trans. on Power Apparatus and Systems,
obtain the pdf of the approximated tcc. Vol. 99, 1980, pp. 195-206
3) Calculate P{system being transient stable} based on the [6] George J. Anders, “Probability Concepts in Electric Power Systems”,
John Wiley & Sons, 1990
pdf of the approximated tcc and the fault clearing time. [7] Arthur R. Bergen, “Power Systems Analysis”, Prentice-Hall Inc.,
4) Repeat the above steps when the time of next run comes. 1986
[8] Hsiao-Dong Chiang, Felix F. Wu and Pravin P. Varaiya,
Table 3. Simulation Times “Foundations of the Potential Energy Boundary Surface Method for
Power System Transient Stability Analysis” IEEE Transactions on
Simulation Times (second) Circuits and Systems, Vol. 35, No. 6, pp. 712-728, June 1988
Model Case 1 Case 2 [9] Linyan Xu, C.O. Nwankpa and Robert Fischl, “Stochastic
Robustness in Dynamic Power System Analysis”, Proceedings of the
26th North American Power Symposium, Part II, pp. 696-701,
Exact 1146 1150
Manhattan, Kansas September 26-27, 1994
[10] J. M. Hammersley and D.C. Handscomb, “Monte Carlo Methods”,
John Wiley & Sons, 1964
Linearly approximated 13.69 13.85 [11] Athanasios Papoulis, “Probability, Random Variables, and Stochastic
Processes”, McGraw-Hill, Inc., 1991
[12] Ronald G. Wasley and Walter O. Stadlin, “Network Applications in
Energy Management Systems” IEEE Computer Applications in
Power, Vol. 4 Number 1, pp. 31-36, January, 1993
[13] Saleh Aboreshaid, Roy Billinton and Mahmud Fotuhi-Firuzabad,
“Probabilistic Transient Stability Studies Using the Method of
VI. CONCLUSIONS Bisection”, IEEE Trans. on Power systems, Vol. 11, No. 4, November
1996, pp. 1990-1995
In this paper, the necessity of modeling tcc as a random
variable in power system transient stability analysis is APPENDIX
addressed. For the purpose of real time on line application,
linear approximation method is proposed to reduce the
G (x + ∆x ) = ∫ f (t , x + ∆x )dt
a
(A-1)
complexity and time for obtaining the pdf of tcc. It is pointed x + ∆x
dt cc let t = z + ∆x , we have
out that the key issue in this method is the calculation of . a − ∆x
dPm G (x + ∆x ) = ∫ f (z + ∆x , x + ∆x )dz (A-2)
x
Expanding f ( z + ∆x , x + ∆x ) around (z , x), we obtain the
A new formula is derived to overcome the difficulty of
calculating this derivative. Monte Carlo Simulation method is
applied to verify the accuracy of the proposed method. Results following equation,
from two examples show that the approximation is accurate f ( z + ∆x , x + ∆x )
under the condition of small load deviation. Methods of
increasing the accuracy of the proposed approximation method
 ∂f ( z , x ) ∂f ( z , x ) 
= f (z , x ) + 
 ∂ z
+
∂ x 
( )
 ∆x + h(z , x )O ∆x (A-3)
2

are suggested.
Substituting (A-3) into (A-2) and doing some manipulations,
we have equation (A-4) as follow.
ACKNOWLEDGEMENT
G( x + ∆x ) = ∫ f (z , x )dz − ∫ f (z , x )dz
a a
x a − ∆x
The authors thank the NSF under Project Number ECS-
9453407 for the funding of this project.

0-7695-1435-9/02 $17.00 (c) 2002 IEEE 7


Proceedings of the 35th Hawaii International Conference on System Sciences - 2002

+∫
( ) + ∂f (z , x )  ∆x + h(z , x )O(∆x 2 )dz (A-4)
a − ∆x   ∂ f z , x BIOGRAPHIES
 
x
 ∂z ∂x   Yiqiao Liang Received his B.S. and M.S. in electrical engineering
Based on the definition of G(x) and equation (A-4), we can from Zhejiang University, P.R. China in 1984 and 1987 respectively.
obtain the following equation. Since 1987, he worked in the System operation department and the
G (x + ∆x ) − G (x ) Communication&automation department of Wenzhou Electric Power
f (z , x )dz
1 a
∆x ∫a −∆x
=− Company Wenzhou, China. From 1991 to 1992 he worked for Lee
∆x
( )
County Electric Cooperative, INC., Florida as an engineer. Currently,
a − ∆x  ∂f ( z , x ) ∂f (z , x )  O ∆x 2 a −∆x
h(z , x )dz
he is a graduate student in ECE department at Drexel University.
+∫
∆x ∫x
 ∂z + dz +
x  ∂x  Saffet Ayasun (M’97) is currently pursuing his Ph.D. degree in
(A-5) Electrical and Computer Engineering Department at Drexel
According to the definition of derivative, we know that University, Philadelphia, PA. His research interests include stability
dG (x ) G (x + ∆x ) − G (x ) of the large-scale nonlinear dynamical system, nonlinear control
= lim (A-6) theory, power systems, and bifurcation theory and power electronics.
dx ∆x →0 ∆x
a  ∂f (z , x ) ∂f (z , x ) 
C.O. Nwankpa received his Magistr Diploma in Electric Power
Since ∫  + dz and f (a, x ) exist, we Systems from Leningrad Polytechnical Institute, USSR, in 1986. He
x  ∂z ∂x  received his Ph.D. degree in the Electrical and Computer Engineering
have the following equations. Department at the Illinois Institute of Technology in 1990. He is
currently an Associate Professor in the Electrical and Computer
∫ f (z , x )dz = f (a , x )
1 a
lim (A-7) Engineering Department at Drexel University. He is the recipient of
∆x →0 ∆x a − ∆x a 1991 NSF Engineering Research Initiation Award and a 1994
 a − ∆x  ∂f (z , x ) ∂f ( z , x )   Presidential Faculty Fellow Award. His research interests are in the
lim  ∫  ∂z + ∂x dz  areas of power systems and power electronics.
∆x →0 x   
a  ∂f ( z , x ) ∂f ( z , x ) 
=∫  + dz (A-8)
x  ∂z ∂x 

lim
( )
O ∆x 2 a −∆x
∫x h(z , x )dz = 0 (A-9)
∆x →0 ∆x
Applying equation (A-6) to equation (A-5) and knowing
equations (A-7)-(A-9), we can obtain the equation below.
dG (x ) a  ∂f (z , x ) ∂f (z , x ) 
= − f (a , x ) + ∫  + dz (A-10)
dx x  ∂z ∂x 
This ends the derivation of the formula.

0-7695-1435-9/02 $17.00 (c) 2002 IEEE 8

You might also like