Vous êtes sur la page 1sur 10

1592 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 29, NO.

4, JULY 2014

Generation and Transmission Expansion Planning:


MILP–Based Probabilistic Model
Jamshid Aghaei, Member, IEEE, Nima Amjady, Senior Member, IEEE, Amir Baharvandi, and
Mohammad-Amin Akbari

Abstract—This paper describes a new probabilistic model Set of all prospective transmission lines.
for generation and transmission expansion planning (G&TEP)
problem considering reliability criteria. Probabilistic reliability Set of all prospective generating units.
criteria accounts for random generator or line outages with
known historical forced outage rates (FOR). The resultant model Set of all existing and newly installed
considers the installation and operation costs as well as the cost of components in the system, i.e., generation
expected energy not supplied (EENS) to optimally determine the units and transmission lines.
number and location of new generating units and circuits in the
network, power generation capacity for those units and the voltage Set of failed components corresponding to
phase angle at each node. Also, efficient linear formulations are contingency .
introduced in this paper to deal with the nonlinear nature of the
problem including objective functions and constraints. Modified
6-bus test system, IEEE 24-bus RTS and IEEE 118-bus test B. Constants
system are utilized to illustrate the effectiveness of the proposed
framework.
Investment cost of constructing line ($).
Index Terms—Expected energy not supplied (EENS), generation
and transmission expansion planning (G&TEP), loss of load prob- Investment cost of new unit ($).
ability (LOEP), mixed integer linear programming (MILP), mixed Operation and maintenance cost of generating
integer nonlinear programming (MINLP).
unit ($/MWh).
Size of the th generating unit (MW).
NOMENCLATURE
Power demand in the demand interval
The main notation used throughout this paper is stated below (MW).
for quick reference. Other symbols are defined as needed
throughout the paper. Duration of demand interval (hour).
Capacity of line (MW).
A. Indices and Sets
Susceptance of line .
Set of generating units located at bus .
Sending-end bus of line .
Set of demands located at bus .
Receiving-end bus of line .
Set of both prospective and existing generating Forced outage rate.
units.
Set of demands.
C. Variables
Set of demand intervals.
Binary variable that is equal to 1 if line is
Set of all networks buses. built and 0 otherwise.
Set of all transmission lines, prospective and Binary variable that is equal to 1 if unit is
existing. constructed and 0 otherwise.
Power output of generator during the demand
Manuscript received February 20, 2013; revised June 01, 2013, September interval (MW).
10, 2013, and November 18, 2013; accepted December 19, 2013. Date of pub-
lication January 09, 2014; date of current version June 16, 2014. Paper no. Power flow through line in the demand
TPWRS-00212-2013. interval (MW).
J. Aghaei, A. Baharvandi, and M.-A. Akbari are with the Department
of Electrical and Electronics Engineering, Shiraz University of Tech- Voltage angle of bus in the demand interval .
nology, Shiraz, Iran (e-mail: aghaei@sutech.ac.ir; a.baharvandi@sutech.ac.ir;
m.akbari@sutech.ac.ir). Involuntary load shedding of th demand
N. Amjady is with the Department of Electrical Engineering, Semnan Uni- during demand interval due to contingency
versity, Semnan, Iran (e-mail: amjady@semnan.ac.ir).
(MW).
Digital Object Identifier 10.1109/TPWRS.2013.2296352

0885-8950 © 2014 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.
AGHAEI et al.: GENERATION AND TRANSMISSION EXPANSION PLANNING: MILP–BASED PROBABILISTIC MODEL 1593

Probability of no contingency state, i.e., all the network is presented and formulated based on mixed integer
components are available. linear programming (MILP) approach. Additionally, [15]
presents G&TEP problem, which incorporates resource ade-
Probability of contingency .
quacy assessment into the optimization framework, solved by a
Expected energy not supplied by th stochastic MILP method. Another stochastic G&TEP research
demand during the demand interval due to work is presented in [16], which uses Monte-Carlo simulation
contingency (MWh). method to model random outages of the equipment and applies
MILP approach for solving the problem [16]. The work of [17]
Some of these variables include subscript , which refers to
copes with generation and transmission expansion investments
contingency .
in Southern Africa for a 20-year planning horizon through LP
approach.
I. INTRODUCTION Reliability is one of the most important factors considered

T HE generation and transmission expansion planning


(G&TEP) problem is a problem of determining the
following major objectives [1]:
in power system planning and operation in both vertically
integrated and deregulated utility environments. Usually,
measurement of reliability or adequacy to make sure that the
1) When to invest? generation and transmission facilities are sufficient to meet
2) How much capacity to add? the future system demand requirements can be done by two
3) What type of generation or transmission is needed? common approaches: 1) deterministic, and 2) probabilistic. The
4) Where to locate new transmission lines and generating most common deterministic criterion is the planning reserve
units? margin (PRM) that can be set equal to the largest online gener-
The solution to the G&TEP problem is to find the most eco- ating unit capacity in the system, as in the Ontario system [18],
nomical expansion plan to decide whether to invest in new gen- or to some fraction of the peak demand as in the Spanish system
eration, new transmission, or a combination of both of them. [19]. An important drawback of the deterministic criteria is
The main purpose of G&TEP is ensuring certain reliability level that they cannot consider the stochastic nature of the system’s
for the forecasted electricity demand, while the generation and behavior. Deterministic analysis using just reserve margin
transmission constraints are satisfied [1]–[3]. calculation could lead to over-investment or insufficient system
In [4]–[6], the generation expansion planning (GEP) problem reliability in the G&TEP. Therefore, it is logical to analyze
is studied. In [4], two objective functions, i.e., cost and envi- such systems based on the probabilistic techniques.
ronmental impacts, are considered to solve a multiobjective System adequacy based on probabilistic criteria is measured
GEP problem. Similarly, in [5] the impact of some of the most directly by stochastic methods such as Monte Carlo simulation.
popular incentive systems on generation planning is considered. To tackle such stochastic problem, the relatively complex math-
In [6], an integrated GEP model towards low-carbon economy ematical and computational techniques should be applied [20],
is proposed and also in order to simplify the proposed model, a such that some of these techniques were unavailable until the
compromised modeling approach is presented. In [7] and [8], last decade or two. Furthermore, it is common knowledge that
transmission expansion planning (TEP) problem is studied. G&TEP problem is a large-scale mixed integer nonlinear pro-
The authors in [7] propose a TEP framework in a market gramming (MINLP) problem, which its nonlinearity is origi-
environment in which only the generation sector is deregulated. nated from the setting of nonlinear constraints such as reliability
Also, [8] is dealt with TEP problem considering the interactions metrics. To get the optimal solution, each possible combination
between large-scale wind integration and transmission system of applicable options along a planning period must be exam-
planning. In [9], the G&TEP problem is presented. The method ined, which leads to a computational explosion in the G&TEP
of [9] introduces a static planning method, which models the problem [21].
least cost deviation from the initial point. The possibility of customers being disconnected for any
Also, some research works cope with G&TEP problem reason can be reduced by increasing the investment cost during
through mixed integer linear programming (MILP) approaches. the planning phase. Over investment can lead to excessive
In [10], a MILP-based multi-objective method is proposed costs. On the other hand, under investment can lead to low
for solving G&TEP problem simultaneously minimizing two reliability level. Trade-off between these two aspects is a major
objectives including: 1) the cost of operation and capital cost, challenge to power system managers, planners, designers,
and 2) emission. In [11], G&TEP problem is solved over a and operators. Several research works consider loss of load
specific planning horizon for an electric utility system so that probability (LOLP) to cope with the reliability issue [22]–[24].
the operation and investment costs would be minimized. A In these papers, reliability is considered as a constraint, which
decomposition approach is applied to reduce the computation should be satisfied.
time of G&TEP problem in [12]. In this reference, the problem In this paper, the EENS due to the random outages of gen-
is solved by linear programming (LP). The research work of erating units and transmission lines is considered as an extra
[13] models the trade-off between generation and transmission objective function. In the earlier studies [22]–[24], EENS and
investment and a bi-level model is proposed for this purpose. LOLP are calculated based on nonlinear formulation. Due to
Accordingly, this model is converted to a single-level mixed complexity of the nonlinear reliability formulation, a linear
integer linear problem. In [14], a three-level equilibrium model model that is compatible with the MILP solution tools is
for the generation and transmission expansion of an electric proposed to calculate the reliability criterion. To the best of
1594 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 29, NO. 4, JULY 2014

the authors’ knowledge, the proposed linear G&TEP model is this line is set to zero. Additionally, constraints (9) and (10) en-
specific to this paper and has not been presented in the previous force the line flow limits. Finally, constraints (11) and (12) en-
research works in the area. Consequently, the corresponding force the angle limits.
G&TEP problem can be solved by MILP methods through one
of the high performance commercially available solvers like B. Expected Cost of Load Shedding
CPLEX and OSL [25], [26]. System reliability is the second objective of the proposed
The remainder of this paper is organized as follows: The model, which should be maximized. This goal can be achieved
mathematical formulation is proposed in Section II. Section III by minimizing the amount of EENS. For calculating the EENS
comprises the formulation and linearization of MOP. Section IV reliability metric at load demand , time interval under contin-
presents numerical experiments and discussions. Some relevant gency , the following formulations are used:
conclusions are drawn in the Section V.
(13)
II. MATHEMATICAL FORMULATION
In this section, the G&TEP problem is formulated as an op-
timization model in which the objective function (comprising
the cost of operation and installation as well as EENS), is min-
imized while satisfying the system constraints.

A. Cost of Operation and Installation (14)


The investment cost for new transmission lines and new gen- (15)
erating units as well as the operation cost of existing and newly (16)
installed generating capacity should be minimized which is cal-
(17)
culated as follows [27], [28]:
(18)

Constraint (13) limits the generation of units to their capacity


limit. Sum of the loads, which is shed from the th demand at
(1) bus during the demand interval , is defined by the nodal bal-
where is the cost of installation and is the cost of ance equality constraint (14). Load shedding is limited by (15).
generation in the case of no contingency occurrence. Constraints on transmission lines are imposed by (16)–(18).
The constraints are listed in the following: Suppose that the state of system components is

(2) (19)
(3)
where denotes the state of existing components that is equal
(4) to 1 and denotes the state of the new units and lines, which
(5) may be constructed or not, and therefore maybe equal to one
(6) or zero. Considering two possible states for system components
(lines and units), available or unavailable, and assuming that the
random variables are statistically independent lead to the proba-
bility definitions based on Bernoulli distribution [29]. The prob-
(7) abilities of no contingency and single contingency occurrence
are as shown in (20) and (21), respectively:
(8)
(20)
(9)
(10)
(11)
(12)

Constraints (2) and (3) denote that existing lines and gener-
ating units have already been built. Constraints (4) and (5) are (21)
binary variables declarations. Constraint (6) establishes the size
of generating units. Constraint (7) enforces the power balance where denotes the forced outage rate of component . Note
at each bus. Constraint (8) represents the power flow through a that (20) and (21) explicitly describe the probability of the
specific line. Note that in (8), is multiplied random outage events in terms of the G&TEP variables, using
by a binary variable , thus, if the corresponding line is not nonlinear formulation. Here, non-contingent and single con-
physically connected to the network, the power flow through tingent states are considered due to their higher probabilities.
AGHAEI et al.: GENERATION AND TRANSMISSION EXPANSION PLANNING: MILP–BASED PROBABILISTIC MODEL 1595

However, higher order contingencies (e.g., double contingen- B. Equivalent Linear Formulation of the Cost Function
cies) can be similarly formulated and incorporated into the Substituting (21) in renders
model if required.
Now, it is possible to express the EENS at load demand , in
time interval and during contingency in terms of the proba-
bility, and amount of load shedding in the contingency:

(22)

Thus, the cost of EENS, denoted by , which is consid-


(27)
ered as the second part of the objective function, is computed
as follows:
where
(23)

(28)
where VOLL is the value of lost load considered as a constant
[30].

C. Probabilistic Model of G&TEP Assume that the product of the binary variable and positive
variable in the second term of (27) equals to . It can be
With consideration of the probabilistic reliability criterion, linearized based on the following linear inequalities:
the new G&TEP model becomes a probabilistic programming
framework. Accordingly, the objective function is given by (29)
(24):
(30)

(31)

Now, another variable is introduced to linearize the


product of and in (27):
(24)

The term of the objective function does not depend on VOLL. (32)

III. EQUIVALENT LINEAR FORMULATIONS


The above formulation is nonlinear due to the product of vari- (33)
ables, e.g., in (8), (21) and (22). This probabilistic MINLP for-
mulation can be transformed into a probabilistic MILP model
by the equivalent set of linear equalities and inequalities. (34)
A. Equivalent Linear Formulation of the Line Flow (35)
Constraints
Nonlinear constraints (8)–(10) can be replaced by the equiv- (36)
alent linear ones as follows:
where is devoted to existing components
(25) of the system and is
devoted to new components; NE and is the number of ex-
isting and new components, respectively. In the first place, it is
(26) assumed that the product value of availability of existing units
, i.e., , is calculated as
where is the disjunctive parameter that should be equal to a a constant. Secondly, to account for the status of the candidate
sufficiently large positive constant. We can see that if equals components , each of
to zero, must vanish due to (25) while the bounds in (26) are them should be considered one by one to calculate .
inactive. Otherwise, when equals to 1, from (26), must be In (32), the probability of availability of existing components
equal to ), while its absolute value is limited to is considered and also, to consider the probability of availability
the line capacity from (25). The same procedure can be applied of new components, (33) and (34) are applied. If a new unit
to the linearization of (16)–(18). or line is not constructed, then the value of corresponding to
1596 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 29, NO. 4, JULY 2014

that unit or line is equal to 0, then would be equal to


due to (33) and (34) would be inactive. On the contrary, when a
specific unit or line is constructed, the value of corresponding
to that unit or line is equal to 1, then would be equal to
due to (34) while the bounds in (33) are
inactive. Accordingly, the last , including all the terms, is equal
to .

C. Equivalent Linear Formulation of the EENS Objective Fig. 1. Single line diagram of six-bus test system.
Function
By substituting (21) in (23) the cost of EENS can be reformu- TABLE I
lated as follows: TRANSMISSION LINE DATA FOR IEEE 6-BUS TEST SYSTEM

(37)

where

(38)

suppose that

(39) IV. NUMERICAL RESULTS AND DISCUSSION


All simulations of this paper are implemented within the
Then, the equivalent linear relations of are as follows:
GAMS software package [31] on a Pentium IV, 2-GHz Core i7
(40) with 8 GB RAM. Three case studies including IEEE six-bus,
RTS and 118-bus systems are considered in this section.
(41)
A. Six-Bus Test System
(42) The six-bus test system is depicted in Fig. 1. The data for
existing and candidate lines is shown in Table I[32]. Similarly,
Thus, nonlinear function of can be linearized by a proce- the data for existing and candidate generators is given in
dure similar to (32)–(35) as follows: Table II[32]. For the sake of simplicity, the planning horizon is
assumed to include only one year. The study period is divided
to five segments with five load factors. For each segment, the
(43) product of its load factor and the annual peak load is considered
as the load of the segment.Table III gives the data corresponding
to the load factors. In the simulation results, the share of buses
3, 4 and 5 in the total load is 0.4, 0.3 and 0.3, respectively.
(44) Also, the value of VOLL is assumed to be $1000/MWh.
There are two typical studies in this section. At first, the
G&TEP model without considering contingencies are analyzed
(45) (Case A). Secondly, the effect of contingencies and reliability
measure are included leading to the proposed probabilistic
(46)
G&TEP model (Case B).
Finally, the objective function can be substituted by linear 1) Case A: G&TEP Model Without Considering Contingen-
formulation as follows: cies: In this case, all of the contingencies are ignored. Thus,
the terms of objective function corresponding to the contin-
(47) gencies are omitted. In fact, in this condition, and
AGHAEI et al.: GENERATION AND TRANSMISSION EXPANSION PLANNING: MILP–BASED PROBABILISTIC MODEL 1597

TABLE II TABLE V
GENERATOR DATA FOR IEEE 6-BUS TEST SYSTEM RESULTS OF G&TEP MODEL IN CASE A

TABLE VI
RESULTS OF GEP MODEL IN CASE B

TABLE III
LOAD FACTOR DATA FOR IEEE 6-BUS TEST SYSTEM

TABLE IV
RESULT OF GEP MODEL IN CASE A
TABLE VII
RESULTS OF G&TEP MODEL IN CASE B

which means all of the system’s components are avail-


able. Therefore, the objective function just consists of the first
two terms of (24) and equals to . To have better in-
sight on the importance of implementing the proposed G&TEP,
in this case the results of GEP have been presented, too. It is
noted that, in the GEP model, line flow limits have been con-
sidered. The results of the GEP and G&TEP models in this case
are shown in Tables IV and V, respectively. According to these
tables, as the value of annual peak load increases, the number
of constructed components increases in order to supply the load
and thus the value of the total cost is increased, too. Likewise,
the GEP method can supply the maximum annual peak load of
53.328 MW. Hence, as the Table IV shows, when the value of
annual peak load is equal to 55 MW, the GEP model would be framework. Indeed, in the GEP model, limited capacity of trans-
infeasible. On the other hand, the G&TEP model can supply the mission lines does not allow to supply more loads. Moreover, in
higher annual peak load of 84.14 MW due to possibility of con- the first two rows of Tables IV and V, which have the same an-
structing both new units and transmission lines in the G&TEP nual peak load, the total cost of both GEP and G&TEP models
1598 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 29, NO. 4, JULY 2014

TABLE VIII
COMPARISON BETWEEN LINEAR AND NONLINEAR FORMULATIONS IN CASE B

are equal because in these states, no line is constructed in the


G&TEP model and thus its total cost is equal to the obtained
value in the GEP model. However, in the rows 3–5, the total
cost of G&TEP model is less than GEP model. Its reason is that
the GEP model can supply the future load by only adding more
units to the system, while the proposed G&TEP model can opti-
mize the system expansion plan considering both generation and
transmission expansion alternatives. When there is enough gen-
erating capacity in the system, the G&TEP model can supply the
future demand by adding transmission lines in the appropriate
places, avoiding from constructing unnecessary units, leading
to more economic solutions for the system expansion plan.
For instance, Tables IV and V show that in the case of 50-MW
annual peak load, GEP model results in constructing 6 new units Fig. 2. Effect of increasing VOLL on the EENS.
with a total cost of $ versus lines with
a total cost of $ in the G&TEP model. These ben- TABLE IX
efits illustrate the effectiveness of the proposed G&TEP model CANDIDATE LINES FOR IEEE 24-BUS TEST SYSTEM
compared to GEP model.
2) Case B: Probabilistic Model Considering the Effect of
Contingencies and Reliability Measure: In this case, the effect
of contingencies on both the operation costs and reliability level
of the system is considered, which leads to the objective func-
tion introduced in (24). In this case, all of the proposed lineariza- any simplifying assumption and compromising accuracy, which
tion techniques are applied. The results of GEP and G&TEP means that there is no approximation leading to decreasing the
models are shown in Tables VI and VII, respectively. Similar accuracy of the model. In other words, we neither ignore nor
to previous section, the GEP model can supply the maximum approximate the nonlinear terms of the original MINLP model.
annual peak load of 53.328 MW and the G&TEP can supply Instead, the proposed linear model, including linearized forms
the maximum of 84.14 MW. In the annual peak loads of 30, 35 of the nonlinear terms, by means of its constraints enforces the
and 40 MW, the GEP and G&TEP models lead to the same total variables so change that the same results of the original non-
costs. However, for the load levels of 45 MW and 50 MW, the linear model are obtained. Consequently, all parts of the pro-
proposed G&TEP model results in lower total costs compared posed linear model produce the same results of the original non-
to the GEP model and the difference increases for the higher linear model without any approximation. Thus, we can say that
annual peak load of 50 MW. In comparison with the previous the proposed linear model is equivalent to the nonlinear model.
case, in the presence of reliability measure to derive a reliable However, the proposed MILP model of G&TEP problem can be
expansion plan, more components with higher costs should be solved much more effectively and better solutions with lower
constructed. This matter can be seen from Tables IV and VI as computation times can be found for it compared to the original
well as Tables V and VII. To show the effectiveness of the pro- MINLP form.
posed linear formulation, the results of the original nonlinear In the real planning environment, we usually have limitation
G&TEP model and the proposed linear framework are presented on the amount of the budget. As shown in Fig. 2, in the presence
in Table VIII. At first, it is seen that the nonlinear model cannot of 55 MW annual peak load and the maximum allowable cost of
solve the G&TEP problem for the annual peak loads greater than $ , as the value of VOLL increases, the value of EENS
35 MW and the simulation results become infeasible. On the decreases. This matter comes from this fact that the value of
other hand, the linear formulation can solve the problem and the objective function should be kept less than its maximum
find optimum feasible solution for the higher load levels up to limit. As shown in Fig. 2, EENS decreases with sharp slop when
84.14 MW. Table VIII also shows that, even for the load levels VOLL is in the range between $100/MWh and $1000/MWh.
that the original nonlinear model can solve the G&TEP problem, However, by further increasing VOLL from $
the proposed linear formulation results in lower total costs with to $ , the EENS slightly decreases as additional
lower computation times. This is due to the fact that the pro- improvement of the system reliability in the high reliability
posed MILP form of the G&TEP problem is obtained without levels is a hard and costly task. This numerical experiment
AGHAEI et al.: GENERATION AND TRANSMISSION EXPANSION PLANNING: MILP–BASED PROBABILISTIC MODEL 1599

TABLE X
COMPARISON BETWEEN LINEAR AND NONLINEAR G&TEP MODELS ON IEEE 24-BUS RTS

TABLE XI
RESULTS OF THE PROPOSED LINEAR G&TEP MODEL FOR IEEE 118-BUS TEST SYSTEM

illustrates the concept of reliability cost versus reliability worth model can obtain considerably lower cost. Moreover, the linear
for the planning studies. and nonlinear formulations solve this problem in 1 min and 37
s and 17 min and 13 s, respectively. Thus, the linear model has
B. IEEE 24-Bus Test System significantly lower computation time than the nonlinear model.
The data of generating units and transmission lines for IEEE It is seen that the advantages of the proposed linear G&TEP
24-bus RTS system is given in [33]. The test system consists model are more highlighted for the larger test system of RTS.
of 32 generators and 17 constant-power type loads. In this test
system, the value of VOLL is assumed to be $1000/MWh as the C. IEEE 118-Bus Test System
previous test case. Also, the data for load factors are as men- IEEE 118-bus test system includes 54 units, 186 branches and
tioned in Table III. 91 loads. Data of this test system can be found in [34] and [35].
In this case, all the generating units except hydro units are In this case existing generators and lines are considered as can-
considered as the candidate generators. Also, the data for the didate ones. The value of VOLL is $1000/MWh. The load fac-
candidate lines are shown in Table IX. For the sake of concise- tors are assumed to be as mentioned for the previous test sys-
ness, only the G&TEP results are presented for this test system tems. The results obtained from the proposed linear G&TEP
in Table X and the results of the linear and nonlinear models are model for IEEE 118-bus test system are shown in Table XI. The
compared. For all load levels of Table X, the proposed MILP value of annual peak load changes from 10 000 to 14 000 MW
model leads to lower total costs and computation times than and the list of newly constructed generators and transmission
the original MINLP model. For instance, in the case of 6000 lines for each annual peak load are given in this table. Similar
MW annual peak load, the value of the objective function for to the previous case studies, when the value of annual peak load
the linear and nonlinear G&TEP models is equal to increases, the number of constructed components increases in
$ and $, respectively, which shows that the linear order to supply the extra load. As a comparison, the original
1600 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 29, NO. 4, JULY 2014

TABLE XII V. CONCLUSION


COMPARISON BETWEEN THE PROPOSED MILP MODEL AND 10 OTHER
METHODS FOR THREE PLANNING HORIZONS OF 15-UNIT TEST SYSTEM This paper deals with generation and transmission expansion
planning and to enhance the reliability of the system, EENS cri-
terion has been incorporated into the model. In the literature,
less attention has been paid to G&TEP problem, compared to
GEP and TEP, due to its complexity. In this research work, the
nonlinear formulation of the G&TEP problem is transformed
to a linear model to cope with the computational complexity
of G&TEP without compromising the modeling accuracy. To
account for the uncertain behavior of the components, contin-
gencies are also modeled in the proposed approach. Effective-
ness of the proposed G&TEP model compared to more conven-
tional GEP model is illustrated. Furthermore, the effectiveness
of the suggested linear formulation in comparison with the orig-
inal nonlinear form based on the ability for finding feasible solu-
tions, optimality of results and computation time is shown. Also,
the effect of contingencies and reliability issues on the G&TEP
results is studied.
nonlinear model could not solve the G&TEP problem on IEEE
118-bus test system, even for the lowest annual peak load of REFERENCES
Table X, due to large solution space and computational com-
[1] A. Lopez, K. Ponnambalam, and V. H. Quintana, “Generation and
plexity of this test case. On the other hand, the proposed linear transmission expansion under risk using stochastic programming,”
model can easily solve with low computation times the G&TEP IEEE Trans. Power Syst., vol. 22, no. 3, pp. 1369–1378, Aug. 2007.
problem on IEEE 118-bus test system for all annual peak loads [2] P. Murugan, S. Kannan, and S. Baskar, “Application of NSGA-II al-
gorithm to single-objective transmission constrained generation expan-
of Table XI. This comparison clearly illustrates the computa- sion planning,” IEEE Trans. Power Syst., vol. 24, no. 4, pp. 1790–1797,
tional efficiency of the proposed linear G&TEP model. Nov. 2009.
[3] I. Sharan and R. Balasubramanian, “Integrated generation and trans-
mission expansion planning including power and fuel transportation
D. 15-Unit Test System constraints,” Energy Policy, vol. 43, pp. 275–284, Jan. 2012.
[4] J. L. C. Meza, M. B. Yildirim, and A. S. M. Masud, “A model for
the multiperiod multiobjective power generation expansion problem,”
This test case includes 15 existing and 18 candidate gen- IEEE Trans. Power Syst., vol. 22, no. 2, pp. 871–878, May 2007.
erating units from five different types. Its technical data is [5] F. Careri, C. Genesi, P. Marannino, M. Montagna, S. Rossi, and I.
taken from [21]. On this test system, the proposed MILP Siviero, “Generation expansion planning in the age of green economy,”
IEEE Trans. Power Syst., vol. 26, no. 4, pp. 2214–2223, Nov. 2011.
model is compared with 10 other solution approaches, which [6] Q. Chen, C. Kang, Q. Xia, and J. Zhong, “Power generation expansion
consist of the numerical optimization method of dynamic planning model towards low-carbon economy and its application in
programming (DP) and nine computational intelligence based china,” IEEE Trans. Power Syst., vol. 25, no. 2, pp. 1117–1125, May
2010.
methods including evolutionary programming (EP), simulated [7] A. Motamedi, H. Zareipour, M. O. Buygi, and W. D. Rosehart, “A
annealing (SA), tabu search (TS), evolutionary strategies (ES), transmission planning framework considering future generation expan-
genetic algorithm (GA), ant colony optimization (ACO), par- sions in electricity markets,” IEEE Trans. Power Syst., vol. 25, no. 4,
pp. 1987–1995, Nov. 2010.
ticle swarm optimization (PSO), hybrid approach (HA) and [8] Y. Gu, J. D. McCalley, and M. Ni, “Coordinating large-scale wind inte-
differential evolution (DE) [36], [37]. In [36] and [37], GEP gration and transmission planning,” IEEE Trans. Sustain. Energy, vol.
problem is solved by these methods for three planning horizons 3, no. 4, pp. 652–659, Oct. 2011.
[9] B. Alizadeh and S. Jadid, “Reliability constrained coordination of gen-
of short-term (6-year), long-term (14-year) and very long-term eration and transmission expansion planning in power systems using
(24-year). For the sake of a fair comparison, the same problem, mixed integer programming,” IET Gener. Transm. Distrib, vol. 5, no.
planning horizons and test system data are also considered 9, pp. 948–960, 2011.
[10] H. Tekiner, D. W. Coit, and F. A. Felder, “Multi-period multi-objective
for the proposed model. The results and computation times electricity generation expansion planning problem with Monte-Carlo
obtained from these 10 methods, quoted from [36] and [37], simulation,” Electr. Power Syst. Res., vol. 80, no. 12, pp. 1394–1405,
and the proposed MILP model are presented in Table XII. The Dec. 2010.
[11] R. M. Sawey and C. D. Zinn, “A mathematical model for long range ex-
computation times of the other methods are measured on a pansion planning of generation and transmission in electric utility sys-
Pentium personal computer in [36] and [37], which is similar tems,” IEEE Trans. Power App. Syst., vol. PAS-96, no. 2, pp. 657–666,
to the hardware set employed in this research work. Table XII Mar. 1977.
[12] M. Pereira, L. Pinto, S. Cunha, and G. C. Oliveria, “A decomposition
shows that DP cannot solve this test case for very long-term approach to automated generation/transmission expansion planning,”
planning horizon and ACO cannot solve both long-term and IEEE Trans. Power App. Syst., vol. PAS-104, no. 11, pp. 3074–3083,
very long-term cases, while the proposed MILP model can Nov. 1985.
[13] M. Jenabi, S. F. Ghomi, and Y. Smeers, “Bi-level game approaches for
solve all three cases. Moreover, the proposed model obtains coordination of generation and transmission expansion planning within
more optimum results (lower costs) with lower computation a market environment,” IEEE Trans. Power Syst., vol. 28, no. 3, pp.
times than all 10 other methods for all three planning horizons. 2639–2650, Aug. 2013.
[14] D. Pozo, E. E. Sauma, and J. Contreras, “A three-level static MILP
These comparisons further illustrate the effectiveness of the model for generation and transmission expansion planning,” IEEE
proposed approach. Trans. Power Syst., vol. 28, no. 1, pp. 202–210, Feb. 2013.
AGHAEI et al.: GENERATION AND TRANSMISSION EXPANSION PLANNING: MILP–BASED PROBABILISTIC MODEL 1601

[15] A. M. Rudkevich, “A nodal capacity market for co-optimization of gen- [33] “Reliability test system task force, the IEEE reliability test system
eration and transmission expansion,” in Proc. Communication, Con- 1996,” IEEE Trans. Power Syst., vol. 14, no. 3, pp. 1010–1020, Aug.
trol, and Computing (Allerton), 2012 50th Annual Allerton Conf., 2012, 1999.
pp. 1080–1088. [34] Power Systems Test Case Archive [Online]. Available: http://www.ee.
[16] M. Pantoš, “Stochastic generation-expansion planning and diversifica- washington.edu/research/pstca.
tion of energy transmission paths,” Electr. Power Syst. Res., vol. 98, [35] IEEE 118 Bus Test System [Online]. Available: http://motor.ece.iit.
pp. 1–10, May 2013. edu/Data
[17] B. Graeber, R. Spalding-Fecher, and B. Gonah, “Optimising trans-na- [36] S. Kannan, S. Slochanal, P. Subbaraj, and N. P. Padhy, “Application
tional power generation and transmission investments: a Southern of particle swarm optimization technique and its variants to generation
African example,” Energy Policy, vol. 33, no. 18, pp. 2337–2349, expansion planning problem,” Electr. Power Syst. Res., vol. 70, no. 3,
Dec. 2005. pp. 203–210, Dec. 2004.
[18] Independent Electricity Operator. Market Rules independent Elec- [37] S. Kannan, S. M. R. Slochanal, and N. P. Padhy, “Application and com-
tricity Operator. Toronto, ON, Canada, Apr. 2002 [Online]. Avail- parison of metaheuristic techniques to generation expansion planning
able: http://www.theimo.com/imoweb/manuals/marketdocs.asp problem,” IEEE Trans. Power Syst., vol. 20, no. 1, pp. 466–475, Feb.
[19] Red Eléctrica de Espana. Operating del Sistema Eléctrico, Proced- 2005.
imientos de operacion, Red Eléctrica de Espana. Madrid, Spain,
Aug. 1998 [Online]. Available: http://www.ree.es/index_sis.html
[20] M. Milligan and B. Parsons, “A comparison and case study of capacity
credit algorithms for intermittent generators,” in Proc. Solar, 97, Wash- Jamshid Aghaei (M’12) received the B.Sc. degree in electrical engineering
ington, DC, USA, Apr. 27–30, 1997. from Power and Water Institute of Technology, Tehran, Iran, in 2003 and the
[21] P. Jong-Bae, P. Young-Moon, W. Jong-Ryul, and K. Y. Lee, “An im- M.Sc. and Ph.D. degrees from Iran University of Science and Technology,
proved genetic algorithm for generation expansion planning,” IEEE Tehran, in 2005 and 2009, respectively.
Trans. Power Syst., vol. 15, no. 3, pp. 916–922, Aug. 2000. His research interests include renewable energy systems, smart grids, elec-
[22] J. Choi, T. Tran, A. A. El-Keib, R. Thomas, H. Oh, and R. Billinton, tricity markets, and power system operation and restructuring.
“A method for transmission system expansion planning considering Dr. Aghaei is a member of the Iranian Association of Electrical and Electronic
probabilistic reliability criteria,” IEEE Trans. Power Syst., vol. 20, no. Engineers.
3, pp. 1606–1615, Aug. 2005.
[23] G. J. Anders, “Generation planning model with reliability constraint,”
IEEE Trans. Power App. Syst., vol. PAS-100, no. 12, pp. 4901–4908,
Dec. 1981. Nima Amjady (SM’10) was born in Tehran, Iran, on February 24, 1971. He re-
[24] S.-L. Chen, T.-S. Zhan, and M.-T. Tsay, “Generation expansion plan- ceived the B.Sc., M.Sc., and Ph.D. degrees in electrical engineering from Sharif
ning of the utility with refined immune algorithm,” Electr. Power Syst. University of Technology, Tehran, Iran, in 1992, 1994, and 1997, respectively.
Res., vol. 76, pp. 251–258, Jan. 2006. At present, he is a Professor with the Electrical Engineering Department,
[25] The ILOG CPLEX website 2006 [Online]. Available: http://www.ilog. Semnan University, Semnan, Iran. Also, he is a consultant with regional electric
com/products/cplex/ companies in Iran. His research interests include power system operation and
[26] GAMS website 2006 [Online]. Available: http://www.gams.com/ planning, load and price forecasting, and artificial intelligence and its applica-
solvers/solvers.html tions to the problems of power systems.
[27] J. L. C. Meza, M. B. Yildirim, and A. S. M. Masud, “A multiobjective
evolutionary programming algorithm and its applications to power gen-
eration expansion planning,” IEEE Trans. Syst., Man, Cybern. A: Syst.
Humans, vol. 39, no. 5, pp. 1086–1096, Sep. 2009.
[28] K. W. Hedman, M. C. Ferris, R. P. O’Neill, E. B. Fisher, and S. S. Amir Baharvandi was born in Iran in 1988. He received the B.Sc. degree
Oren, “Co-optimization of generation unit commitment and transmis- from Shahid Chamran University of Avaz, Iran, in 2011, and the M.Sc. degree
sion switching with N-1 reliability,” IEEE Trans. Power Syst., vol. 25, from Shiraz University of Technology, Shiraz, Iran, in 2013, both in electrical
no. 2, pp. 1052–1063, May 2010. engineering.
[29] F. Partovi, M. Nikzad, B. Mozafari, and A. M. Ranjbar, “A stochastic His research interests are power system operation, planning, and reliability.
security approach to energy and spinning reserve scheduling consid-
ering demand response program,” Energy, vol. 36, pp. 3130–3137, Apr.
2011.
[30] T. Limbu, “Value-based allocation and settlement of reserves in Mohammad-Amin Akbari was born in Iran in 1985. He received the B.Sc.
electricity markets,” IET Gener. Transm. Distrib., vol. 5, no. 4, pp. degree from Hormozgan University, Iran, in 2009, and the M.Sc. degree from
489–495, 2011. Shiraz University of Technology, Shiraz, Iran, in 2013, both in electrical engi-
[31] Generalized Algebraic Modeling Systems (GAMS) [Online]. Avail- neering. He is currently pursuing the Ph.D. degree at the Department of Elec-
able: http://www.gams.com tronics and Electrical Engineering, Shiraz University of Technology, Shiraz,
[32] R. J. Hyung, M. Shahidehpour, and W. Lei, “Market-based genera- Iran.
tion and transmission planning with uncertainties,” IEEE Trans. Power His research interests are smart grids, power system operation and planning,
Syst., vol. 24, no. 3, pp. 1587–1598, Aug. 2009. reliability, and optimization methods.

Vous aimerez peut-être aussi