Beruflich Dokumente
Kultur Dokumente
a r t i c l e i n f o a b s t r a c t
Article history: This paper presents a method in expansion planning of transmission systems using the AC optimal power
Received 21 October 2013 ow (AC-OPF). The AC-OPF provides a more accurate picture of power ow in the network compared to
Received in revised form 28 March 2014 the DC optimal power ow (DC-OPF) that is usually considered in the literature for transmission expan-
Accepted 16 April 2014
sion planning (TEP). While the AC-OPF-based TEP is a mixed-integer nonlinear programming problem,
Available online 8 May 2014
this paper transforms it into a mixed-integer linear programming environment. This transformation guar-
antees achievement of a global optimal solution by the existing algorithms and software. The proposed
Keywords:
model has been successfully applied to a simple 3-bus power system, Garvers 6-bus test system, 24-bus
AC-OPF
Linearized power ow
IEEE reliability test system (RTS) as well as a realistic power system. Detailed case studies are presented
Mixed-integer linear programming and thoroughly analyzed. Simulations show the effectiveness of the proposed method on the TEP.
Transmission expansion planning 2014 Elsevier B.V. All rights reserved.
1. Introduction presented in [10] for the TEP which has been tested on a realistic
power system. In [11], a scenario-based multi-objective model is
Transmission expansion planning (TEP) addresses the problem presented for multi-stage TEP where the non-dominated sorting
of augmenting transmission lines of an existing transmission net- genetic algorithm (NSGA-II) is used to overcome the difculties
work; the objective is to optimally serve a growing electric load in solving the non-convex mixed-integer optimization problem. A
while satisfying a set of economical, technical and reliability con- multi-objective framework is presented in [12] for the TEP in dereg-
straints [1]. In general, the TEP is considered as making a stochastic ulated environment. Recently, intermittent energy resources have
decision on when (the time), where (the location), and which types been experiencing a rapid growth in power generation around the
of transmission lines to be installed. In [2,3], a classication scheme world. Therefore, new challenges are introduced to integrate the
categorizes the subjects published in this area. renewable energy sources (RES) to the power grid. There are many
A mixed-integer linear programming (MILP) is used in [4], published papers which focus on this newly main issue [1315].
for the TEP problem that considers power losses. We suggest The planner of power system should deal with many uncer-
a linearization method for the AC-TEP based on the method in tainties during the planning process such as load uncertainty,
[4]. However, both active and reactive powers are included in uncertainty in prices, market rules and etc. A multistage TEP prob-
our proposed formulation. In [5], a multi-year TEP model is pre- lem including available transfer capability (ATC) is modeled in
sented using a discrete evolutionary particle swarm optimization [16] that takes load uncertainty into account by considering sev-
approach. Aguado et al. [6], present a novel TEP model that con- eral scenarios generated by Monte Carlo simulation. In [17], the
siders a multi-year planning horizon in a competitive electricity TEP is studied by considering the load uncertainty using ben-
market. Both the TEP and generation expansion planning (GEP) ders decomposition. Since this paper is focused on transforming
problems are analyzed together in [7,8]. Also, transmission switch- a mixed-integer nonlinear programming (MINLP) problem into
ing (TS) is investigated in [9] for the TEP showing that the TS could a MILP, uncertainties have not been considered in this paper.
improve the capacity expansion planning model as well as reduc- However, the presented model can easily be extended for taking
ing the total planning cost. A meta-heuristic and holistic approach is uncertainties into account.
The surveyed literatures use the DC-OPF for solving the TEP
problem which is not completely suitable due to ignoring reactive
power. However, there have been some published papers which
Corresponding author. Tel.: +98 0912 5179374; fax: +98 21 88462066. use the AC-OPF to solve the TEP problem [1821]. This paper pro-
E-mail addresses: tohidakbari@yahoo.com, tohidakbari@ee.kntu.ac.ir (T. Akbari). poses an approach for transmission planning based on the AC-OPF,
http://dx.doi.org/10.1016/j.epsr.2014.04.013
0378-7796/ 2014 Elsevier B.V. All rights reserved.
94 T. Akbari, M. Tavakoli Bina / Electric Power Systems Research 114 (2014) 93100
Nomenclature +
ijt , ijt positive variables used so as to eliminate the abso-
Indices lute function
g index of generators it voltage angle at bus i in time t
i, j indices of buses
l index of lines
m index of blocks used for piecewise linearization
t index of sub-periods (times) providing a more accurate picture of both active and reactive
power ows in the expanded power network in the future planning
Sets horizon. The novelty of this paper is the introduction of a MILP for-
B set of all buses mulation using the AC-OPF approach so as to solve the expansion
CL set of all candidate lines planning problem of transmission grid. In brief, an AC-OPF-based
CLi set of all candidate lines connected to bus i TEP is formulated, and linearized around the operating point in
EL set of all existing lines order to derive a MILP problem. Solving a MILP problem is a mature
ELi set of all existing lines connected to bus i technology, where the MILP solvers can be embedded in many tools
G set of all generators and applications. Moreover, some numerical examples are pre-
SP set of all sub-periods sented in which simulations are discussed accordingly. The whole
linearization process provided in subsection B of section II that
Constants converts the non-linear AC approach to a MIP problem is novel and
max
APL.I maximum apparent power ow of line l has not been previously presented. This paper contributes to the
I interest rate TEP by approximating the sine and cosine functions in power ow
ICl investment cost of candidate line l equations by their Taylors series; then, the quadratic function is
M number of blocks used for piecewise linearization modeled using piecewise linear functions. Moreover, the inequal-
nc number of candidate lines ity constraints for apparent powers of existing and candidate lines
ng number of generators are transformed into a set of linear constraints. Numerical results
PD,it active power demand at bus i in time t conrm the contribution of the proposed method in comparison
QD,it reactive power demand at bus i in time t with the conventional solutions. Since the proposed optimization
min
PG,g minimum active power of generator g problem for solving the TEP is linear, the global optimal solution
max
PG,g maximum active power of generator g can be obtained easily by the available software. In addition, out-
min
comes obtained by the proposed method are more accurate (due
QG,g minimum reactive power of generator g
to taking reactive power into account) than those of the available
max
QG,g maximum reactive power of generator g conventional methods (due to ignoring reactive power by using
|Vimin | minimum of the voltage magnitude at bus i the DC-OPF for solving the TEP). It should be emphasized that the
|Vimax | maximum of the voltage magnitude at bus i ISO (independent system operator) is responsible for transmission
Yij0 , Yij admittance of line ij for the existing and candidate expansion planning; the ISO aims at minimizing the investment
lines, respectively. (Yij0 = Gij0 + jBij0 , Yij = Gij + jBij ) cost plus the total payment to the generating companies.
ij,m slope of the mth block of the voltage angle difference
between corridor i and j 2. Analysis and formulation of the TEP
ij maximum of each block width for corridor i and j
ref voltage phase angle for the slack bus ( ref = 0) Here the proposed method is formulated, presenting the TEP
a constant to make investment and operation cost based on the AC-OPF using a MINLP that will be transformed into a
comparable MILP.
ij , , ij , 1 , ij , disjunctive parameters
2.1. The AC-OPF-based TEP formulation
Variables:
ICt investment cost in time t The objective function is the investment cost (IC) which is the
OCt operation cost in time t construction cost of new lines and transformers (if any) plus the
OF objective function operation cost (OC). The OC includes the total cost of generation in
PG,gt active power of generator g in time t the power system under study. Using an AC power ow, objective
PL0 ,lt active power ow of existing line l from bus i to bus function of the TEP can be formulated as follows:
ij
j in time t
PLij , lt active power ow of candidate line l from bus i to
bus j in time t
nc
ng
Min OF = (1 + I)
t ult ICl + Cg PG,gt (1)
QG,gt reactive power of generator g in time t
QL0 ,lt reactive power ow of existing line l from bus i to tSP
l=1 g=1
ij
bus j in time t Investment cost in time t(ICt ) Operation cost in time t(OCt )
QLij , lt reactive power ow of candidate line l from bus i to
bus j in time t The objective function (1) is subjected to the following equality and
ult binary variable related to the candidate lines in time inequality constraints;
t Equality constraints:
|Vit | voltage magnitude at bus i in time t
ij,mt width of the mth angle block of corridor i and j in PG,it PD,it = PL0 + PLij ,lt i, j B, t SP
lELi ij ,lt lCLi
time t
(2)
T. Akbari, M. Tavakoli Bina / Electric Power Systems Research 114 (2014) 93100 95
QG,it QD,it = QL0 + QLij ,lt i, j B, t SP MINLP formulation (hardly attainable and no guarantee to reach
lELi ij ,lt lCLi it), the MILP solution remains close to the global solution of origi-
(3) nal MINLP problem depending on the number of blocks in modeling
the circle or quadratic terms (M or n).
PL0 = |Vit |2 Gij0 |Vit ||Vjt | (Gij0 cos(it jt ) 2.2. Linearization of the presented MINLP
ij ,lt
+ Bij0 sin(it jt )) i, j B, l EL, t SP (4) Here it is introduced the linearization of the MINLP, presented
by (1)(12), around the normal operating point. Let us assume
the phase difference between the voltages at both ends of every
PLij ,lt = ult (|Vit |2 Gij |Vit ||Vjt | (Gij cos(it jt ) existing or constructed lines is small enough; this implies vali-
dation of rst order approximation of sine and cosine functions
+ Bij sin(it jt ))) i, j B, l CL, t SP (5) in their Taylor series. Moreover, assume the voltage magnitudes
are nearly 1 p.u. for all buses. These assumptions are practically
true under normal operating condition, maintaining the system far
QL0 = |Vit |2 Bij0 |Vit ||Vjt | (Gij0 sin(it jt ) from instability and other security limits. Based on the aforemen-
ij ,lt
tioned assumptions, (4)(7) can be re-arranged by substituting sine
+ Bij0 cos(it jt )) i, j B, l EL, t SP (6) and cosine functions in their Taylor series expansion about zero.
In the following formulation, just for the sake of simplicity, we
have dropped the subscript t. It can be added to the corresponding
QLij ,lt = ult (|Vit |2 Bij |Vit ||Vjt | (Gij sin(it jt )) variables or parameters if referring to sub-period t.
+ Bij cos(it jt ))) i, j B, l CL, t SP (7) PL0 = Gij0 (1 cos(i j )) Bij0 sin(i j )
ij ,l
2
(i j )
Inequality constraints:
= Gij0 Bij0 (i j ) i, j B, l EL (14)
2
min
PG,g max
PG,gt PG,g g G, t SP (8)
PLij ,lt = u1 (Gij (1 cos(i j )) Bij sin(i j ))
min
QG,g QG,gt max
QG,g g G, t SP (9)
2
2 2 max 2 (i j )
(PL0 ,lt ) + (QL0 ,lt ) (APL,l ) l EL, t SP (10)
= u1 Gij Bij (i j ) i, j B, l EL (15)
ij ij 2
2
(PLij ,lt )2 + (QLij ,lt )2 ult (APL,l
max
) l CL, t SP (11)
QL0 = Bij0 (1 cos(i j )) Gij0 sin(i j )
|Vimin | |Vit | |Vimax | i B, t SP (12) ij ,lt
2
u1(t+1) u1t l CL, t SP (13) (i j )
= Bij0 Gij0 (i j ) i, j B, l EL (16)
2
While (1) shows the objective function, i.e. IC plus OC, (2) and
(3) represent active and reactive power balance at each bus, respec-
tively. Constraints (4) and (5) indicate active power owing through QLij ,l = u1 (Bij (1 cos(i j )) Gij sin(i j ))
existing and candidate lines, respectively. Similarly, constraints (6)
2
and (7) introduce reactive power ows through existing and can- (i j )
didate lines, respectively. Superscript index 0 is used to denote
= u1 Bij Gij (i j ) i, j B, l CL (17)
2
the existing lines. Active and reactive power generation limits of
the generators are represented by (8) and (9). Transmission power
limits are shown by (10) and (11) for both the existing and can- Then, the second order functions in (14)(17) can be linearized
didate lines, respectively. Voltage magnitude limits are shown by by considering 2M piecewise linear blocks as shown in Fig. 1,
(12). Eq. (13) implies that if a candidate line is constructed within approximating the existing line l between corridors i and j as below:
the sub-period t, it is considered as an existing line in the next
M
sub-period. (i j ) =
2 0
ij,m ij,m , i, j B (18)
The above-formulated problem is a MINLP problem due to the
m=1
nonlinearities in (4)(7) as well as in (10)(11). It should be noted
that the obtained solution for the model (1)(12) is a practical feasi-
M
ble local optimum solution that meets the requirements of ISO; but, |i j | = 0
ij,M , i, j B (19)
not a global one. This is due to the non-convexity nature of the for- m=1
mulated problem. However, when one tries solving a non-convex 0
0 ij,m ij , i, j B (20)
MINLP optimization problem, there will be no guarantee to obtain
the global optimum solution. This restriction remains unresolved ij,m = (2m 1) ij m = 1, . . ., M i, j B (21)
for the most practical optimization models in a complex system
0
where ij,m and ij,m are the slope and the value of the mth block of
[22,23].
Hence, the proposed MINLP problem is transformed into an phase difference between corridors i and j, respectively. The slope
MILP to avoid any local optimal solutions. This obtained MILP for- can be calculated according to (21). The same constraints will be
mulation is an approximate for the original AC formulation. Further, applied to the candidate lines. However, (20) should be rewritten
the global optimal solution of the MILP formulation can efca- for candidate lines as follow:
ciously be obtained by existing solvers. While the resultant solution
is not necessarily the same as the global optimal solution of the 0 ij,m ij + (1 u1 )ij i, j B (22)
96 T. Akbari, M. Tavakoli Bina / Electric Power Systems Research 114 (2014) 93100
360 k 360
cos cos (k 1) QL0
n n ij ,l
360
max
APL,l sin 0 k = 1, . . ., n, l EL (30)
n
360 k 360
u1 sin sin (k 1) PLij ,l
n n
360 k 360
cos cos (k 1) QLij ,l
n n
360
Fig. 1. Piecewise linear modeling of loss function.
max
APL,l sin 0, k = 1, . . ., n, l CL (31)
n
Appropriate values for ij and ij can be selected as below: The proposed model has been successfully applied to four differ-
ent power systems; a simple 3-bus power system, Garvers 6-bus
ij = 2
2 Gij 2
Bij (28) test system, IEEE 24-bus reliability test systems (RTS) and Iranian
ij = 2
Gij 2
Bij 2
(29) 400-kV network. This problem was solved on a PC running with
Core 2 Duo CPU, clocking at 2.00 GHz and 1 GB RAM. The used
It should be emphasized that the above linearization tech- software is CPLEX 12.2.0.2 under GAMS 23.6.3 (General Algebraic
nique adds three sets of continuous variables to the problem, Modeling System) [24]. Cplex is a commonly used solver to solve
0 ( +
ij,m ij,m ), ij and ij (M + 2 variables for each line between cor- the MIP problems which can efcaciously get to the global optimal
ridors i and j). Further, nonlinearity in (10) can be removed by solution. This solver uses the branch and cut algorithm.
T. Akbari, M. Tavakoli Bina / Electric Power Systems Research 114 (2014) 93100 97
15 candidate branches, and a total demand of 760 MW. Generators, Number of blocks of 26 Number of single 19,483
loads and lines data can be found in [4]. Reactive power demand is equations equations
assumed to be 20% of active power demand for each bus. Maximum Number of blocks of 14 Number of single 7018
variables variables
Number of non-zero 41,872 Number of discrete 15
Table 1 elements variables
Simulation data for the 3-bus test system.
Table 4
Candidate lines data.
Candidate lines Capacity (MVA) Reactance (p.u.) Investment cost ($106 US)
Name From To
Table 5
Results for different value of .
lines are operating at their capacity limits. The number of non-zero regular polygon is used to model the circle (see Appendix for more
elements and single equations are 60,304 and 28,699, respectively. details).
Other simulation data for the AC-OPF and the DC-OPF are the same The parameter was set to be 25 8760. The execution time
as those of listed in Table 2. Execution time reported by the GAMS was 3.32 seconds. Simulations show for the LAC-OPF that lines 8,
was about 0.296 s. 10 and 11 should be built within the planning horizon. Generating
Third, for n = 64 in the LAC-OPF, the objective function will units 2, 10, 11, 14, 19, 20, 31 and 32 will produce active powers
become $4.34 108 and candidate lines to be built were 2, 3, 5, 6, of 29.4 MW, 145.5 MW, 0 MW, 12.7 MW, 15.6, 0 MW, 216.9 MW
9, 10, 11, 12, 14 and 15. For n > 64, simulations remain unchanged. and 0 MW, respectively; other units will operate at their maxi-
Therefore, one should be careful in determining the number of seg- mum capacities. Table 5 indicates the results for various , showing
ments used to model the circle. Using the AC-TEP will result in the considerable changes in results when varies. Using a DC-TEP
construction of the same lines as those of the LAC-TEP under n = 64 approach for this case shows that lines 8 and 10 are constructed.
with the exception of building line 13 instead of line 12; the objec- This can be justied as reactive power is neglected in the DC
tive function for the AC-OPF becomes $5.29 108 that is bigger than approach, and therefore more transmission line capacities are
that of the LAC-OPF. available; resulting in lower required investment cost. Also, in an
Assuming = 10 8760,1 the IC and OC are equal to $4.34 108 AC-TEP approach the candidate lines 5, 8, 9 and 10 are selected to
and $17,398.7 per hour for n = 64. This means the same candi- be built which explicitly shows that the obtained solution is sub-
date lines are constructed when considering operation cost. The optimal due to the non-convexity nature of the AC-OPF equations.
Lagrangian multiplier of active power balance equation or loca-
tional marginal price (LMP) for bus numbers 16 are listed in 3.4. Fourth case study: Iran 400-kV simplied network
Table 3.2 It should be highlighted that these LMPs are obtained
after xing the binary variables of candidate lines to their corre- As the last test, a realistic power system is chosen; i.e., Iranian
sponding optimal values of the optimization problem. Note that 400-kV power system. The topology and data of the network can be
the generator connected across bus 6 is the marginal generator. found in [11]. Power factor of loads is set to be 0.9 as a typical value.
Table 3 shows LMPs are different at those buses since power losses Instead of considering a single period as in previous case studies, a
are not ignored. It should also be noted that there is no congestion dynamic approach is selected for this case. To do so, an interest rate
in transmission lines. of 8% is regarded. Impedances of three-bundled and two-bundled
400-kV overhead lines are considered to be 0.0192 + j0.288 /km
and 0.0288 + j0.3296 /km, respectively. Also, annual load growth
3.3. Third case study: the IEEE 24-bus RTS is considered to be equal to 5% in average. The system is comprised
of 52 buses, 28 generating units and 27 candidate lines. A planning
The well-known 24-bus RTS is used to apply the presented horizon of 16 years is considered which is divided into two sub-
method. Parameters of the candidate lines are listed in Table 4, periods.
while resistances of lines are assumed to be one fth of their cor- The problem was run for different values of n. Simulations have
responding reactances. Detailed data and topology of the RTS can been reported in Table 6. As it can be seen from the obtained results,
be found in [25]. In order to put the transmission lines under more the consumed times to reach the optimal solution will increase sig-
stress, active power demands and maximum active power gener- nicantly by increasing the number n. To reduce the solution time,
ation capacities are multiplied by 2 for each generator. A 32-sided the number n can be different for various lines. It clearly depends on
the operating point of active and reactive power for each line in the
corresponding PQ plane. The number n have to be big for highly
1
congested and heavily loaded lines; but, n can be small without
Please note that 8760 is the number of hours per year.
2
It is to be noted that to calculate LMPs, the objective function has been modied
losing any generality for the lines operating far from their lim-
to include only the generation offers (or equivalently operating cost) i.e. investment its. Solving the TEP using simple DC power ow might be helpful
cost must be removed from the aforementioned objective function. to identify the most congested lines or areas. In the meanwhile,
T. Akbari, M. Tavakoli Bina / Electric Power Systems Research 114 (2014) 93100 99
Table 6
Results from different cases for Iran 400-kV power system.
M = 10 3,910,760/1,828,279/54 3.38
n=8
M = 10 5,122,152/1,828,279/54 3.72
n = 16
M = 10 7,544,936/1,828,279/54 6.43
n = 32
M = 10 12,390,504/1,828,279/54 43.14
n = 64
Table 7
Optimal solutions for Iran 400-kV power system.
[2] C.W. Lee, K.K. Ng, J. Zhong, F. Wu, Transmission expansion planning from past [14] A.M. Leite da Silva, L. Manso, W. Sales, S.A. Flvio, G.J. Anders, Chronolog-
to future, Public Saf. Commun. Eur. (2006) 257265. ical power ow for planning transmission systems considering intermittent
[3] G. Latorre, R. Cruz, J.M. Areiza, A. Villegas, Classication of publications and sources, IEEE Trans. Power Syst. 27 (4) (2012) 17221733.
models on transmission expansion planning, IEEE Trans. Power Syst. 18 (2) [15] Y. Gu, J.D. McCalley, M. Ni, Coordinating large-scale wind integration
(2003) 938946. and transmission planning, IEEE Trans. Sustain. Energy 3 (4) (2012) 652
[4] N. Alguacil, A.L. Motto, A.J. Conejo, Transmission expansion planning: a mixed- 659.
integer LP approach, IEEE Trans. Power Syst. 18 (3) (2003) 10701077. [16] T. Akbari, A. Rahimikian, A. Kazemi, A multi-stage stochastic transmission
[5] M.C. Da Rocha, J.T. Saraiva, A multiyear dynamic transmission expansion plan- expansion planning method, Energy Convers. Manage. 52 (2011) 2844
ning model using a discrete based EPSO approach, Electr. Power Syst. Res. 93 2853.
(2012) 8392. [17] T. Akbari, S. Zolfaghari, A. Kazemi, Multi-stage stochastic transmission expan-
[6] J.A. Aguado, S. Torre, J. Contreras, A.J. Conejo, A. Martinez, Market-driven sion planning under load uncertainty using benders decomposition, Int. Rev.
dynamic transmission expansion planning, Electric Power Syst. Res. 82 (1) Electr. Eng. 4 (2009) 976984.
(2012) 8894. [18] M.J. Rider, A.V. Garcia, R. Romero, Power system transmission network expan-
[7] B. Graeber, Generation and transmission expansion planning in southern Africa, sion planning using AC model, IET Generation, Transm. Distrib. 1 (5) (2007)
IEEE Trans. Power Syst. 2 (1999) 983988. 731742.
[8] M. Kandil, S. El-Debeiki, N. Hasanien, Rule-based system for determining unit [19] M. Rahmani, M. Rashidinejad, E.M. Carreno, R. Romero, Efcient method for
locations of a developed generation expansion plan for transmission planning, AC transmission network expansion planning, Electr. Power Syst. Res. 20 (9)
IEE Proc. Gener. Transm. Distrib. 147 (1) (2000) 6268. (2010) 10561064.
[9] A. Khodaei, M. Shahidehpor, S. Kamalnoa, Transmission switching in expansion [20] R.A. Jabr, Polyhedral formulations and loop elimination constraints for dis-
planning, IEEE Trans. Power Syst. 25 (3) (2010) 17221733. tribution network expansion planning, IEEE Trans. Power Syst. 28 (2) (2013)
[10] V.S.K.M. Balijepalli, S.A. Khaparde, A Holistic approach for transmission system 18881897.
expansion planning studies: an Indian experience, IEEE Trans. Power Syst. 5 (2) [21] T. Akbari, A. Rahimikian, M. Tavakoli Bina, Security-constrained transmis-
(2010) 199212. sion expansion planning: a stochastic multi-objective approach, Electr. Power
[11] P. Maghouli, S. Hosseini, M. Oloomi, M. Shahidehpour, A scenario-based multi- Energy Syst. (IJEPES) 43 (1) (2012) 444453.
objective model for multi-stage transmission expansion planning, IEEE Trans. [22] El.I Samahy, K. Bhattacharya, C. Canizares, M.F. Anjos, J. Pan, A procurement
Power Syst. 26 (1) (2011) 470478. market model for reactive power services considering system security, IEEE
[12] P. Maghouli, S. Hosseini, M. Oloomi, M. Shahidehpour, A multi-objective frame- Trans. Power Syst. 23 (1) (2008) 137149.
work for transmission expansion planning in deregulated environments, IEEE [23] D.P. Bertsekas, Nonlinear Programming, Athena Scientic, NH, 1999.
Trans. Power Syst. 24 (2) (2009) 10511061. [24] Generalized Algebraic Modeling Systems (GAMS), 2013, Available from:
[13] Q. Zhou, L. Tesfatsion, C. Liu, R. Chu, W. Sun, Nash approach to planning mer- http://www.gams.com
chant transmission for renewable resource integration, IEEE Trans. Power Syst. [25] IEEE Committee Report, The IEEE reliability test system 1996, IEEE Trans.
28 (3) (2012) 20862100. Power Syst. 14 (3) (1999) 10101020.