Electrical Power and Energy Systems 53 (2013) 553563

Contents lists available at SciVerse ScienceDirect

Electrical Power and Energy Systems

journal homepage: www.elsevier.com/locate/ijepes

Dynamic stochastic fractional programming for sustainable

management of electric power systems
H. Zhu a, G.H. Huang b,c,
a

Institute for Energy, Environment and Sustainable Communities, University of Regina, Regina, Saskatchewan, Canada S4S 0A2
S-C Institute for Energy, Environment and Sustainability Research, North China Electric Power University, Beijing 102206, China
c
University of Regina, Regina, Canada S4S 0A2
b

a r t i c l e

i n f o

Article history:
Received 9 January 2013
Received in revised form 29 April 2013
Accepted 10 May 2013

Keywords:
Sustainability
Generation expansion planning
Fractional programming
Chance-constrained programming
Renewable energy
Stochastic uncertainty

a b s t r a c t
A dynamic stochastic fractional programming (DSFP) approach is developed for capacity-expansion
planning of electric power systems under uncertainty. The traditional generation expansion planning
focused on providing a sufcient energy supply at minimum cost. Different from using least-cost models,
a more sustainable management approach is to maximize the ratio between renewable energy generation and system cost. The proposed DSFP method can solve such ratio optimization problems involving
issues of capacity expansion and random information. It has advantages in balancing conicting objectives, handling uncertainty expressed as probability distributions, and generating exible capacityexpansion strategies under different risk levels. The method is applied to an expansion case study of
municipal electric power generation system. The obtained solutions are useful in generating sustainable
power generation schemes and capacity-expansion plans. The results indicate that DSFP can support
in-depth analysis of the interactions among system efciency, economic cost and constraint-violation
risk.
 2013 Elsevier Ltd. All rights reserved.

1. Introduction
Sustainable management of energy systems plays a signicant
role in the social and economic development of urban communities. At the present, the major energy sources for electricity generation are non-renewable fossil fuels, which could have serious
consequences for the local and global environment [27]. With the
growing health and environmental awareness of the people, developing renewable energy sources has gained much attention
throughout the world [11,44]. Although numerous optimization
methods have been explored, it is still considered difcult to identify sustainable management plans for hybrid electrical systems.
The rst challenge is to identify a trade-off between conicting
economic and environmental concerns. The second is associated
with uncertainties in the input information, such as future electricity demands and resource availabilities. The third is the reection
of dynamic characteristics of facility capacity issues. Therefore,
efcient mathematical programming techniques for planning the
electric power systems under these complexities are desired.
Corresponding author. Address: S-C Institute for Energy, Environment and
Sustainability Research, Resources and Environmental Research Academy, North
China Electric Power University, Beijing 102206, China. Tel.: +1 391 146 8225;
fax: +1 306 585 4855.
E-mail address: huang@iseis.org (G.H. Huang).
0142-0615/$ - see front matter  2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.ijepes.2013.05.022

Previously, many integrated models were proposed in designing environmentally responsible energy management systems
[43,28,25,34,38,39]. Classically quite a few models were formulated as single-objective linear programming (LP) problems aiming at the minimization of total cost under specic levels of
environmental requirements [23]. For example, Zhu et al. [51]
developed a municipal-scale energy model for the City of Beijing,
where the objective was to minimize the system cost over the
planning horizon, and the constraints included the restrictions
for air pollutant emissions. Occasionally environmental concerns
were directly quantied by economic indicators and encompassed in the aggregated least-cost objective function. For
instance, Li et al. [22] proposed a greenhouse gas (GHG)-mitigation oriented energy system management model, where the
trading of GHG-emission credit was represented through an
economic measure. However, equating the conventional singleobjective least-cost optimization framework to real-world problems involving social and environmental considerations might
lead to difculties in obtaining solutions from a sustainability
perspective.
For better reecting the multi-dimensionality of the sustainability goal, it was increasingly popular to represent the energy
management problems within a Multiple Objective Programming
(MOP) framework [30,31,18,45,1,33]. Thus environmental impacts
were also elected as explicit objective functions in the models

554

H. Zhu, G.H. Huang / Electrical Power and Energy Systems 53 (2013) 553563

besides the least-cost desire [20]. Correspondingly, multi-criteria

decision making (MCDM) approaches, such as goal programming
[36], weighted sum [30,19] and fuzzy multi-objective techniques
[33], were widely employed to determine satisfactory compromise
solutions. In general, the approaches to deal with multi-objective
problems could be classied into two categories: compromise-programming and aspiration-analysis approaches. In compromising
programming, the importance of each objective was delineated
by weighting factors, and the feasible solution with the shortest
overall distance to the ideal values of objectives is considered most
desirable [33]. Thus, the solutions determined by compromising
programming were highly dependent on the preferences of decision-makers. In contrast, the substance of aspiration approaches
was to develop a single objective programming model through
optimizing one objective and converting the others into constraints under certain aspiration levels. This manipulation was
especially practical for large-scale problems; nevertheless, the
trade-offs of multiple objectives were neglected and the system
complexities could not be adequately reected. In general, these
MCDM methods had two major limitations. First, they usually
combined objectives of multiple aspects into a single measure on
the basis of subjective assumptions. The work of setting weighting
factors or economic indicators inevitably entailed additional difculties. Second, they merely focused on system inputs and outputs,
and none of them could facilitate analysis of system efciencies
represented as output/input ratios.
For the optimization of system efciency, Fractional Programming (FP) was used in many management problems
[10,41,42,15], in which the optimization of ratio between two
quantities (e.g. output/input, cost/time, or cost/volume) was desired. The use of ratio objectives in FP problems assures that only
the solutions with better achievements per unit of inputs (e.g. cost,
resource, time) would be selected. In addition, the scenarios where
FP techniques can be applied are the same as those for LP and
MCDM techniques [37,21]. Thus FP can be a natural and powerful
tool for studying and analyzing the issues related to sustainability
in resources management problems. Although applications of FP
were reported in various areas ranging from engineering to economics [32], the method was seldom applied to energy systems
planning. Moreover, few studies of FP under uncertainty were reported [7,17].
In real-world problems of energy systems management, renewable energy resources are normally subject to spatial and/or temporal uctuations; electricity demands are merely imprecisely
estimated [29,2]. The type of uncertainty that attracts major attention is randomness existing in right-hand-side parameters
[14,24,50,49]. For example, the availability of solar and wind energies are uncertain and can be expressed as probability distributions
[4]. The chance-constrained programming (CCP) is an attractive
tool to deal with problems associated with such uncertainties.
Zhu and Huang [48] developed a stochastic linear fractional programming (SLFP) method for supporting sustainable waste management, which incorporated the CCP technique within a FP
framework, and could solve ratio optimization problems associated with random information. However, SLFP was not able to deal
with the capacity expansion issues in electric power systems. One
potential approach to improve SLFP is to integrate the mixed integer linear programming (MILP) technique within its framework.
Therefore, the objective of this study is to develop a dynamic
stochastic fractional programming (DSFP) approach for sustainable
management of electric power systems. The CCP and MILP techniques will be incorporated into a linear fractional programming
(LFP) framework. The integrated DSFP approach can not only deal
with ratio objective, but also reect dynamics of facility expansion
under stochastic uncertainties. The developed method will then be
applied to a case study to demonstrate its advantages. Desired

municipal power system management schemes under different

constraint-violation levels will be obtained, which will help decision makers analyze the interrelationships among renewable
power generation efciency, system risk and many related factors.
2. Methodology
2.1. Mixed integer linear fractional programming
A general linear fractional programming (LFP) problem can be
formulated as follows:

Pn

j1 c j xj

j1 dj xj

Max f x1 ; x2 ; . . . ; xn Pn

s:t:

n
X
aij xj 6 bi ;

i 1; 2; . . . ; m

1a

1b

j1

xj P 0;

j 1; 2; . . . ; n

1c

where aij, bi, cj, dj 2 R; a and b are scalar constants. Assume that the
solution set of model (1) is nonempty and bounded, and the objective function is continuously differentiable.
Charnes and Cooper [8] showed that if the denominator is conPn
stant in sign (assuming that
j1 dj xj b > 0) for all X = (x1, x2,
. . . , xn) on the feasible region, the LFP model can be transformed
to the following linear programming problems under transformation xj r  xj ("j):

Max gx1 ; x2 ; . . . ; xn ; r

n
X
cj xj a  r

2a

j1

s:t:

n
X
aij xj 6 bi  r;

i 1; 2; . . . ; m

2b

j1
n
X
dj xj b  r 1

2c

j1

xj P 0;

j 1; 2; . . . ; n

2d

rP0

2e

Model (2) can be solved through the usual simplex algorithm.

Thus the optimal solution of Model (1) can be obtained through
transformation xj xj =r ("j).
In a mixed integer linear fractional programming (MILFP) problem, some decision variables are dened as integers. Thus the
problem can be formulated as:

Pp

j1 c j xj

Max f Pp

j1 dj xj

s:t:

Pn

jp1 c j yj

jp1 dj yj

Pn

p
n
X
X
aij xj
aij yj 6 bi ;

3b

jp1

j1

xj P 0;

i 1; 2; . . . ; m

3a

j 1; 2; . . . ; pp < n

yj P 0 and yj integer variable; j p 1; . . . ; n

3c
3d

If the denominator is constant in sign (assuming that

P
njp1 dj yj b > 0) on the feasible region, the MILFP
model can be transformed to:

Pp

j1 dj xj

Max g

p
X
j1

cj xj

n
X
jp1

cj yj a  r

4a

555

H. Zhu, G.H. Huang / Electrical Power and Energy Systems 53 (2013) 553563

s:t:

p
n
X
X
aij xj
aij yj 6 bi  r;

i 1; 2; . . . ; m

4b

jp1

j1

j1

p
n
X
X
dj xj
dj yj b  r 1

4c

jp1

j1

yj r  yj ;
xj P 0;

j p 1; . . . ; n

4d

j 1; 2; . . . ; pp < n

4e

yj P 0 and yj integer variable; j p 1; . . . ; n

4f

rP0

4g

Model (4) can be solved through the branch-and-bound algorithm. The solution of integer variables yj (j = p + 1, . . . , n) can be obtained directly, and the optimal solution of xj (j = 1, 2, . . . , p) can still
be obtained through transformation xj xj =r (j = 1, 2, . . . , p).
The MILFP model can tackle ratio optimization problems and
capacity-expansion issues; however, it may not be able to deal
with uncertain parameters.
2.2. Dynamic stochastic fractional programming
When some of the right-hand-side parameters in model (3) are
of stochastic features and can be represented as probability distributions, a dynamic stochastic fractional programming (DSFP) model can be formulated as follows:

Pp

j1 c j xj

Max f Pp

j1 dj xj

s:t:Pr

Pn

jp1 cj yj

jp1 dj yj

Pn

p
n
X
X
aij xj
aij yj 6 bi t P 1  pi ;

xj P 0;

5a

i 1; 2; . . . ; m

5b

jp1

j1

j 1; 2; . . . ; pp < n

yj P 0 and yj integer variable; j p 1; . . . ; n

5c
5d

where t 2 T; bi(t) is a random right-hand-side parameter (in constraint i) dened on a probability space T; pi (pi 2 [0, 1]) is a given
level of probability for constraint i (i.e. signicance level, which represents the admissible risk of constraint violating).
According to the CCP methods [9], when the left-hand-side
coefcients [Ai] are deterministic and the right-hand-side coefcients [bi(t)] are random (for all pi values), the constraint
PrAi X 6 bi t P 1  pi can be converted as:

Ai X 6 bi tpi
pi

p
n
X
X
aij xj
aij yj 6 bi tpi ;

6
F 1
i pi ,

where bi t
given the cumulative distribution function
of bi [i.e. Fi(bi)] and the probability of violating constraint i (pi).
Constraint (6) is linear and the set of feasible constraints is convex. Thus, the CCP method can be used to solve the DSFP model (5)
by converting them into deterministic versions through: (i) xing a
certain level of probability pi (pi 2 [0, 1]) for uncertain constraint i,
and (ii) imposing that constraint should be satised with at least a
probability level of 1  pi .
The detailed solution process for DSFP can be summarized as
follows:
Step 1: Formulate the original DSFP model [i.e. model (5)].
Step 2: Given a signicance level (pi) for each constraint i, formulate the deterministic MILFP model [i.e. model (3)] by converting stochastic constraints [i.e. constraints (5b)] into
deterministic ones:

i 1; 2; . . . ; m

5b

jp1

where bi tpi represents corresponding values given the cumulative distribution function of right-hand-side parameter bi and the
probability of violating constraint i (pi), bi tpi F 1
i pi .
Step 3: Formulate and solve the transformation model [i.e.
model (4)], and obtain solutions for xj (j = 1, 2, . . ., p), yj
(j = p + 1, . . ., n) and r.
Step 4: Calculate the solutions for xj: xj xj =r (j = 1, 2, . . . , p).
Step 5: Repeat steps 24 under different pi levels.
3. Case study
3.1. Overview of study system
The developed DSFP method is applied to a case study of longterm planning of a regional electric power system with representative data within a Chinese context [4,5,26,46]. In the study system,
there is an independent municipal electricity grid, where the electricity can be converted from both fossil fuels (coal and natural gas)
and renewable energies (hydro, solar and wind). With these energy
carriers, ve types of electricity-generation facilities (coal-red,
natural gas-red, hydropower, solar energy and wind power) are
employed by a wholly state-owned electric power company in
the regional power supply system. With the economic development, the total electricity demands of residential, commercial
and industrial end-users would rise accordingly. When the existing
power-generation capacities are insufcient to meet the increasing
energy demands, capacity expansion will become necessary. The
decision makers are responsible for planning the capacity expansions for these facilities and allocating energy resources/services
for electricity supplies through multiple technologies. To reect
the dynamic features of the study system, three time periods (5year for each period) are considered in a 15-year planning horizon.
Currently, the electricity generation in the study system primarily relies on non-renewable fossil fuels, which are only available by extraction or renery with limited supplies. When the
domestic supplies cannot satisfy energy demands, additional
amount of coal and natural gas may be imported with higher
prices. Moreover, carbon dioxide (CO2) from fossil fuels red power
plants is a signicant source of greenhouse gas (GHG) emissions,
which may contribute to climate change and global warming
[47]. Burning fossil fuels also produces other air pollutants, such
as sulfur dioxide (SO2), nitrogen oxides (NOx), volatile organic compounds (VOCs) and heavy metals. In contrast, most of renewable
energy sources are environmentally friendly, and allow the region
to increase its independency on the primary energy supplies. With
the increasing concerns of environmental protection and resources
conservation, the power generation technologies based on renewable energy resources are encouraged. Therefore, the planners desire optimal capacity expansion plans and sustainable resources
allocations, which can achieve maximized renewable power generation with minimized system cost.
Table 1 presents the average market prices for domestic or imported fossil fuels and operation costs for electricity generation in
the three periods. The capital investment costs and capacity expansion options for power-generation facilities are listed in Table 2.
Every facility could expand its capacity once per period, where
three capacity expansion options are provided for each facility.
Once capacity expansions of the facilities are decided for a period,
they would be implemented at the beginning of the period. Thus,
the increased facility capacities could be employed for the current
and the following periods. Table 3 provides electricity demands

556

H. Zhu, G.H. Huang / Electrical Power and Energy Systems 53 (2013) 553563

Table 1
Costs for energy supply and electricity generation.
Time period

Energy supply cost, Pik ($106/PJ)

Coal (i = 1)
Imported coal (i = 2)
Natural gas (i = 3)
Imported natural gas (i = 4)

k=1

k=2

k=3

2.56
3.20
4.89
6.29

2.76
3.50
5.09
6.59

2.96
3.80
5.29
6.89

0.1465
0.5755
0.485
0.00975
1.3745

0.1465
0.5755
0.485
0.00975
1.3745

Electricity generation cost, Cjk ($106/PJ)

Coal (j = 1)
0.1465
Natural gas (j = 2)
0.5755
Hydropower (j = 3)
0.485
Solar (j = 4)
0.00975
Wind (j = 5)
1.3745

Table 2
Capacity expansion options and costs for power-generation facilities.
Time period

Capacity expansion options, Vjmk (GW)

Coal (j = 1)
m=1
m=2
m=3
Natural gas (j = 2)
m=1
m=2
m=3
Hydropower (j = 3)
m=1
m=2
m=3
Solar (j = 4)
m=1
m=2
m=3
Wind (j = 5)
m=1
m=2
m=3

k=1

k=2

k=3

0.04
0.06
0.08
0.012
0.026
0.036
0.006
0.008
0.01
0.004
0.008
0.012
0.006
0.01
0.016

0.04
0.06
0.08
0.012
0.026
0.036
0.006
0.008
0.01
0.004
0.008
0.012
0.006
0.01
0.016

0.04
0.06
0.08
0.012
0.026
0.036
0.006
0.008
0.01
0.004
0.008
0.012
0.006
0.01
0.016

i: Primary energy resources (i = 1, 2, 3, 4);

j: Power-generation facilities (j = 1, 2, . . . , 5);
k: Time periods (k = 1, 2, 3);
m: Capacity-expansion options (m = 1, 2, 3).
For effective management of various energy resources and facilities, these system components and their interactions need to be
considered in an integrated modeling framework. The difculties
of planning this regional power system include: (a) how to indentify optimal capacity-expansion schemes for power-generation
facilities; (b) how to maximize renewable power generation with
a possible low system cost; (c) how to reect the stochastic features of uncertain parameters (e.g. electricity demands, allowable
capacities for solar-energy and wind-power facilities); and (d)
how to analyze the tradeoff between the system efciency and
the risk of system reliability. Thus, the developed DSFP method will
be applied to the planning of such a system. The decision variables
include:
Xjk: Electricity generation at facility j in period k (PJ);
Yjmk: Integer variable (=1 or 0) representing capacity expansion
for facility j with option m in period k will be installed or not;
Zik: Supply of primary energy resource i in period k (PJ).
The objective is to maximize renewable power generation per
unit of system cost, while a series of constraints dene the interrelationships among the decision variables and system conditions/
factors. In detail, the system cost is formulated as a sum of the
following:
(a) costs for primary energy supply:
4 X
3
X

Pik  Z ik

i1 k1

Capital cost, Ejmk ($10 /GW)

Coal (j = 1)
Natural gas (j = 2)
Hydropower (j = 3)
Solar (j = 4)
Wind (j = 5)

771
826
3240
4500
1464

721
776
3040
4300
1364

671
726
2840
4100
1264

(b) xed and variable operating costs for power generation:

5 X
3
X

C jk  X jk

j1 k1

(c) capital (investment) costs for facility expansions:

over the planning horizon, which can be presented as probability
distributions under different pi levels. Table 4 presents the existing
and allowable capacities of power-generation facilities and the
availabilities of energy resources. Due to the natural variations of
renewable energy resources, the distribution information of allowable capacities for solar energy and wind power facilities under different probability levels of constraint violation (pi) is given in
Table 5.

3.2. The DSFP model for regional power systems planning

In the study region, multiple primary energy sources (domestic
and imported production), multiple technologies (conventional
and renewable electricity generation), and multiple facilities exist.
Thus the following indices will be used:

5 X
3 X
3
X

Ejk  V jmk  Y jmk

j1 m1 k1

where
Pik = average cost for primary energy supply i in period k ($106/
PJ);
Cjk = average variable and maintenance cost for generating electricity from facility j in period k ($106/PJ);
Ejk = capital cost for capacity expansion of facility j within period k ($106/GW);
Vjmk = capacity expansion for facility j with option m in period k
(GW).
Thus, the ratio objective of DSFP model can be formulated as
follows:

Table 3
Electricity demands [Dk(t), PJ] in period k under different pi levels.
pi level

0.01

0.05

0.1

0.25

0.5

0.75

0.9

0.95

0.99

D1(t)
D2(t)
D3(t)

9.450
17.715
25.890

12.171
20.441
28.611

13.624
21.894
30.064

16.052
24.322
32.492

18.750
27.020
35.190

21.448
29.718
37.888

23.876
32.146
40.316

25.329
33.599
41.769

28.050
36.325
44.490

557

H. Zhu, G.H. Huang / Electrical Power and Energy Systems 53 (2013) 553563
Table 4
Capacities of power-generation facilities and availabilities of energy resources.
Power-generation facilities

Existing capacity (GW)

Allowable capacity (GW)

Energy resources

Availability (PJ)

Coal (j = 1)
Natural gas (j = 2)
Hydropower (j = 3)
Solar (j = 4)
Wind (j = 5)

0.06
0.04
0.02
0.01
0.01

0.18
0.092
0.036
VC4(t)
VC5(t)

Coal (i = 1)
Imported coal (i = 2)
Natural gas (i = 3)
Imported natural gas (i = 4)

100
60
40
30

Table 5
Allowable capacities (GW) for solar energy and wind power facilities under different pi levels.
pi level

0.01

0.05

0.1

0.25

0.5

0.75

0.9

0.95

0.99

Solar energy, VC4(t)

Wind power, VC5(t)

0.0322
0.0301

0.0334
0.0333

0.0340
0.0350

0.0350
0.0379

0.0362
0.0411

0.0374
0.0443

0.0384
0.0472

0.0390
0.0489

0.0402
0.0521

Table 6
Results of the DFSP model.

Power generation, Xjk (PJ)

Coal-red (j = 1)

Natural gas-red (j = 2)

Hydropower (j = 3)

Solar energy (j = 4)

Wind power (j = 5)

Primary energy supply, Zik (PJ)

Coal (i = 1)

Imported coal (i = 2)

Natural gas (i = 3)

Imported natural gas (i = 4)

Capacity expansion,
Coal-red (j = 1)

P3

m1 V jmk

Natural gas-red (j = 2)

Hydropower (j = 3)

Solar energy (j = 4)

Wind power (j = 5)

Period

pi = 0.01

pi = 0.05

pi = 0.10

pi = 0.25

k=1
k=2
k=3
k=1
k=2
k=3
k=1
k=2
k=3
k=1
k=2
k=3
k=1
k=2
k=3

13.543
13.642
21.153
3.154
7.546
8.199
4.730
5.676
5.676
3.469
4.730
4.730
3.154
4.730
4.730

9.461
14.993
22.075
3.570
3.154
4.242
4.730
5.676
5.676
3.469
4.730
4.730
4.100
5.046
5.046

8.424
12.909
18.922
3.154
3.154
5.311
4.730
5.676
5.676
3.469
5.361
5.361
4.100
5.046
5.046

5.995
9.461
15.768
3.154
3.543
5.406
4.730
5.676
5.676
3.469
5.361
5.361
4.100
5.676
5.676

k=1
k=2
k=3
k=1
k=2
k=3
k=1
k=2
k=3
k=1
k=2
k=3

0
29.982
70.018
44.829
15.171
0
0.321
19.016
20.662
7.626
0
0

0
26.931
73.069
31.315
22.696
0
8.995
7.947
10.689
0
0
0

0
37.369
62.631
27.882
5.360
0
7.947
7.947
13.384
0
0
0

16.493
31.315
52.192
3.352
0
0
7.947
8.929
13.623
0
0
0

k=1
k=2
k=3
k=1
k=2
k=3
k=1
k=2
k=3
k=1
k=2
k=3
k=1
k=2
k=3

0.04
0
0.04
0
0.012
0
0.01
0.006
0
0.012
0.008
0
0.01
0.01
0

0
0.04
0.04
0
0
0
0.01
0.006
0
0.012
0.008
0
0.016
0.006
0

0
0.06
0
0
0
0
0.01
0.006
0
0.012
0.012
0
0.016
0.006
0

0
0
0.04
0
0
0
0.01
0.006
0
0.012
0.012
0
0.016
0.01
0

41.628
67.237
1019.724
40.822

43.204
57.495
878.128
49.200

44.466
51.874
817.760
54.375

45.727
43.327
711.100
64.305

 Y jmk (GW)

Renewable power generation (PJ)

Non-renewable power generation (PJ)
System cost ($106)
Renewable power generation / system cost (PJ per $109)

558

H. Zhu, G.H. Huang / Electrical Power and Energy Systems 53 (2013) 553563

Table 7
Results of the least-cost model.

Power generation, Xjk (PJ)

Coal-red (j = 1)

Natural gas-red (j = 2)

Hydropower (j = 3)

Solar energy (j = 4)

Wind power (j = 5)

Primary energy supply, Zik (PJ)

Coal (i = 1)

Imported coal (i = 2)

Natural gas (i = 3)

Imported natural gas (i = 4)

Capacity expansion,
Coal-red (j = 1)

P3

m1 V jmk

Period

pi = 0.01

pi = 0.05

pi = 0.10

pi = 0.25

k=1
k=2
k=3
k=1
k=2
k=3
k=1
k=2
k=3
k=1
k=2
k=3
k=1
k=2
k=3

13.543
13.642
21.153
3.154
7.546
8.199
4.730
5.676
5.676
3.469
4.730
4.730
3.154
4.730
4.730

9.461
15.768
22.075
3.570
3.640
5.503
4.730
5.676
5.676
3.469
3.469
3.469
4.100
5.046
5.046

8.424
13.540
18.922
3.154
3.154
5.942
4.730
5.676
5.676
3.469
4.730
4.730
4.100
5.046
5.046

5.995
10.481
15.768
3.154
3.154
6.037
4.730
5.676
5.676
3.469
4.730
4.730
4.100
5.676
5.676

k=1
k=2
k=3
k=1
k=2
k=3
k=1
k=2
k=3
k=1
k=2
k=3

0
29.982
70.018
44.829
15.171
0
0.321
19.016
20.662
7.626
0
0

0
26.931
73.069
31.315
25.261
0
8.995
9.173
13.868
0
0
0

0
37.370
62.630
27.882
7.448
0
7.947
7.947
14.974
0
0
0

13.116
34.692
52.192
6.729
0
0
7.947
7.947
15.212
0
0
0

k=1
k=2
k=3
k=1
k=2
k=3
k=1
k=2
k=3
k=1
k=2
k=3
k=1
k=2
k=3

0.04
0
0.04
0
0.012
0
0.01
0.006
0
0.012
0.008
0
0.01
0.01
0

0
0.04
0.04
0
0
0
0.01
0.006
0
0.012
0
0
0.016
0.006
0

0
0.006
0
0
0
0
0.01
0.006
0
0.012
0.008
0
0.016
0.006
0

0
0.04
0
0
0
0
0.01
0.006
0
0.012
0.008
0
0.016
0.01
0

41.628
67.237
1019.724
40.822

40.681
60.017
876.857
46.395

43.204
53.136
816.718
52.900

44.466
44.589
711.069
62.534

 Y jmk (GW)

Natural gas-red (j = 2)

Hydropower (j = 3)

Solar energy (j = 4)

Wind power (j = 5)

Renewable power generation (PJ)

Non-renewable power generation (PJ)
System cost ($106)
Renewable power generation / system cost (PJ per $109)

Max f

renewable power generation

system cost
P5 P3

P4 P3
i1

k1 P ik

 Z ik

P5 P3
j1

j3

k1 C jk

(3) Electricity demand constraints:

k1 X jk

 X jk

P5 P3
j1

m1

P3

k1 Ejk

 V jmk  Y jmk

Pr

5
X
X jk P Dk t P 1  pk;D ;

8k

7e

j1

7a

(4) Energy resources constraints:

The constraints of DSFP are dened as follows:

(1) Mass balance constraints of fossil fuels:

X 1k  CR1 6

2
X

Z ik ;

8k

3
X
Z ik 6 UPi ;

7b

i1
4
X
X 2k  CR2 6
Z ik ;

8k

7c

3 X
k
X
V jmp  Y jmp   Ucap P X jk ;

8j; k

X jk P dj  EC j  Ucap;

8j; k

7g

(6) Capacity expansion constraints:

(2) Capacity constraints for electricity generation:

m1 p1

7f

(5) Technical constraints of electricity generation:

i3

EC j

8i

k1

7d

EC j

3 X
3
X
V jmk  Y jmk 6 VC j ;
k1 m1

j 1; 2; 3

7h

H. Zhu, G.H. Huang / Electrical Power and Energy Systems 53 (2013) 553563

ECj = existing capacity of power-generation facility j (GW);

Ucap = conversion coefcient from power-generation capacity
to energy (GW to PJ);
Dk(t)= total electricity demand in period k (PJ);
UPi = available resource for primary energy i (PJ);
dj = facility utilization rate (%);
VCj = allowable capacity for power-generation facility j (GW);
VCj(t) = allowable power-generation capacity for solar energy
(j = 4) and wind power (j = 5) facilities (GW);
pk,D = constraint-violation probability for electricity demand
constraints;
pj,VC = constraint-violation probability for availability constraints of solar energy (j = 4) and wind power (j = 5).

1050
1000

System cost ($106)

559

950
900
850
800
750

4. Results and discussion

700
p = 0.01

p = 0.05

Optimal-ratio

p = 0.1

p = 0.25

Least-cost

Renewable power generation / cost (PJ / $109)

(A) System cost

65.0

55.0

45.0

35.0
p = 0.01

p = 0.05

Optimal-ratio

p = 0.1

p = 0.25

Least-cost

(B) Renewable power generation per unit of cost

Fig. 1. Comparison of system performances from optimal-ratio and least-cost
models.

(7) Availability constraints of renewable energy resources:

PrEC j

3 X
3
X

V jmk  Y jmk 6 VC j t P 1  pj;VC ; j 4;5

7i

k1 m1

(8) Expansion option constraints:

3
X

Y jmk 6 1;

8j; k

7j

m1

Y jmk 0 or 1;

8j; m; k

7k

(9) Non-negativity constraints:

Z ik P 0;

8i; k

7l

X jk P 0;

8j; k

7m

where
CR1 = converting efciency from coal to electricity;
CR2 = converting efciency from natural gas to electricity;

Table 6 shows the results of the DSFP model under different pi

levels, which include the power generation patterns, primary energy supply strategies and capacity expansion plans. For example,
when pi = 0.01, the electricity generation through coal-red, natural gas-red, hydropower, solar energy and wind power technologies in period 1 would respectively be 13.543, 3.154, 4.730,
3.469 and 3.154 PJ; the primary energy supplies of coal, imported
coal, natural gas and imported natural gas in period 1 would
respectively be 0, 44.829, 0.321 and 7.626 PJ; the capacity-expansion levels for coal-red, natural gas-red, hydropower, solar energy and wind power facilities at the beginning of period 1 would
respectively be 0.04, 0, 0.01, 0.012 and 0.01 GW. Similarly, the
plans for the three periods under different pi levels can be
interpreted.
The results of power generation indicate that renewable electricity supply would be insufcient for the future energy demands.
Over the planning horizon, coal-red electricity would increase
steadily and still play an important role in the power system due
to its high availability and competitive price. In comparison, generating electricity from natural gas is more expensive; thus the natural gas-red facility would only be expanded under some
demanding conditions (i.e., when the electricity demands cannot
be guaranteed by expansions of the other energy generation facilities). For instance, a capacity of 0.012 GW would be added to the
natural gas-red facility at the beginning of period 2 when
pi = 0.01, while no capacity expansion would be taken for this facility under pi = 0.05, 0.1 and 0.25. For the facilities relying on renewable energy resources, capacity expansions would be conducted in
the rst two periods for efcient increase in sustainable electricity
generation. Under various constraint-violation risk levels, the
hydropower facility would not only have similar capacity expansion plans (i.e. expanded by 0.01 GW at the beginning of period
1, and 0.006 GW at the beginning of period 2), but also have similar
electricity generation patterns (i.e. 4.730, 5.676 and 5.676 PJ in
periods 1, 2 and 3). These points of similarity among the energy
management schemes corresponding to different pi levels indicate
that the hydropower-related results are not sensitive to the uncertain system inputs. In contrast, the results related to solar energy
and wind power are very sensitive due to natural variations in
the availabilities of these energy resources.
The DSFP results in Table 6 also reect that a higher ratio
objective and a lower system cost would be obtained under a
higher pi level. For example, when pi is raised from 0.01 to
0.05, the ratio objective would be increased from 40.822 to
49.200 PJ per $109, and the system cost would be decreased
from $1019.724  106 to $878.128  106. The pi level represents
the probability of constraint violation, and the ratio objective
indicates the system efciency (i.e. renewable power generation
per unit of system cost). Thus, the relationship between the
ratio objective and pi can reect the interrelationships among

560

H. Zhu, G.H. Huang / Electrical Power and Energy Systems 53 (2013) 553563

Optimal-ratio model
p = 0.1

p = 0.25

Capacity expansion (GW)

p = 0.05

Least-cost model

k=1

k=2

k=3

k=1

k =2

k=3

k=1

k=2

k=3

Coal-fired facility

Optimal-ratio model
p = 0.1

p = 0.25

Capacity expansion (GW)

p = 0.05

Least-cost model

k=1

k=2

k=3

k=1

k=2

k=3

k=1

k=2

k=3

Solar energy facility

Fig. 2. Comparison of capacity expansion schemes from optimal-ratio and least-cost models.

system efciency, economic cost, and system reliability (i.e.

energy-supply security and constraint-violation risk). A lower
pi level means a lower admissible risk of violating the constraints, which leads to an increased strictness for the constraints and thus a shrunk decision space. When the
availabilities of renewable energy resources are restricted under
a lower pi level, more electricity would be generated from nonrenewable resources to meet the increased electricity demands,
leading to larger capacity expansions for conventional facilities.
Thus an alternative of lower system efciency (i.e. higher system costs and lower renewable power generation) would be obtained, but the reliability of meeting demanding conditions
would also increase. On the contrary, decisions under a higher
pi level would loosen the electricity demands and allow more
power generation from renewable energy resources, leading to
an alternative of decreased reliability but higher system efciency. Therefore, the DSFP approach can provide useful solutions associated with different risk levels, which will help
decision makers identify desired electricity production and
capacity-expansion plans under various system conditions. Generally, planning under a lower pi level would be suitable for
demanding conditions, whereas planning under a higher pi can
be used under advantageous conditions.
Besides the scenario of maximizing the system efciency, another scenario of minimizing the system cost is also analyzed to
evaluate the effects of different energy supply polices. The optimal-ratio problem presented in Model (7) can be converted into
a least-cost problem with the following objective:

Min f system cost

4 X
3
5 X
3
X
X
Pik  Z ik
C jk  X jk
i1 k1

5 X
3 X
3
X

Ejk  V jmk  Y jmk

j1 k1
0

7a

j1 m1 k1

Thus, the obtained least-cost model is a mixed-integer linear

programming (MILP) problem containing stochastic right-hand
parameters, which can still be solved by CCP method. Table 7 presents results of the least-cost model under different pi levels, which
can be compared with those of the optimal-ratio model as shown
in Table 6. When pi = 0.01, the obtained energy management
schemes under these two scenarios are totally identical; however,
when pi takes other values (i.e. pi = 0.05, 0.10 or 0.25), differences
can be found between the results of two scenarios. Obviously,
the optimal solutions corresponding to two different objectives
[i.e. (7a) and (7a0 )] are generally different except when the feasible
solutions dened by the constraints [i.e. (7b)(7m)] are controlled
under an extremely low constraint-violation risk (i.e. the solution
space are severely restricted).
Fig. 1 compares the system performances corresponding to the
optimal-ratio and least-cost scenarios under various pi levels. As
indicated in Fig. 1A, the system costs obtained from these two
models would both decrease with a raised pi level. Although the
least-cost model achieves slightly lower system costs when
pi = 0.05, 0.10 and 0.25, the differences in system cost between
two scenarios are not as obvious as those in system efciency. As
indicated in Fig. 1B, the optimal-ratio model can lead to signicant

H. Zhu, G.H. Huang / Electrical Power and Energy Systems 53 (2013) 553563

561

Fig. 3. Comparison of power-generation patterns from (A) optimal-ratio and (B) least-cost models.

higher system efciencies under these pi levels. For example, when

pi = 0.05, the least-cost model leads to a system cost being 0.145%
lower than the optimal-ratio model; while the optimal-ratio model
results in a system efciency being 6.046% higher than the leastcost model.
Moreover, when pi = 0.05, 0.10 and 0.25, the differences between the obtained results under two energy supply scenarios
are demonstrated in Figs. 2 and 3. Fig. 2 indicates that, even with
the same conditions, a higher capacity of the solar energy facility
would be achieved in periods 2 and 3 under the optimal-ratio scenario. For example, when pi = 0.05, the solar energy facility would
have no capacity expansion in periods 2 and 3 under the least-cost

scenario; whereas it would have the second capacity expansion in

period 2 under the optimal-ratio scenario. In comparison, the coalred facility would have similar capacity expansion schemes under
the two scenarios, except the case when pi = 0.25 (a higher capacity
would be necessary for period 2 under the least-cost scenario).
Accordingly, as shown in Fig. 3, with the same pi level, the optimal-ratio model leads to a relatively higher percentage of solar energy supply; in comparison, the least-cost model leads to relatively
higher percentages of coal-red and natural gas-red electricity
supplies. In addition, these two models lead to same energy supply
schemes for hydropower and wind power facilities. Thus the ratios
of renewable electricity to the total supplied electricity under the

562

H. Zhu, G.H. Huang / Electrical Power and Energy Systems 53 (2013) 553563

optimal-ratio scenario (i.e. 42.90%, 46.16% and 51.35% when

pi = 0.05, 0.10 and 0.25) would be signicantly higher than those
under the least-cost scenario (i.e. 40.40%, 44.85% and 49.93% when
pi = 0.05, 0.10 and 0.25).
The above scenario analysis could help the energy system managers to identify desired electricity supply options under various
electricity supply policies. When the manager aims to increase
renewable power generation while keeping a relatively low system
cost, the DSFP model can be a particularly effective tool for identifying schemes with optimal system efciency under various conditions. The DSFP results can effectively support energy systems
management in (a) balancing conicting objectives, (b) providing
desired capacity-expansion plans, and (c) analyzing the interactions among system efciency, economic cost and energy-supply
reliability. To plan a practical power system expansion, detailed
analytical studies are generally necessary to simulate the operation of the projected system under the future conditions [3]; multicriteria decision analysis (MCDA) techniques [12] would also be
useful for comparing the attributes (e.g. reliability levels, price of
electricity) obtained/estimated from various plans and thus selecting the robust plan.
As the rst attempt to develop a new fractional programming
approach for long-term capacity-expansion planning of electric
power systems under uncertainty, DSFP still has some limitations.
Because additional uncertainties in the objective parameters (e.g.
fuel prices, capital costs) would make the system planning considerably more complex, DSFP only considers the random right-handside parameters in constraints (e.g. electricity demands, renewable
energy resource availability). In the future study, DSFP can be further improved by integrating other uncertain optimization methods within its framework, such as interval programming [16,13]
and fuzzy mathematical programming [40]. Furthermore, when
DSFP is applied to more complicated cases in a competitive electricity market environment, the electricity prices should be input
along the planning horizon, and advanced decision-making techniques such as game theory [6] and genetic algorithms [35] may
need to be incorporated into the modeling frame to deal with the
interaction between the decisions of the different generation
agents.
5. Conclusions
A dynamic stochastic fractional programming (DSFP) approach
has been developed for supporting sustainable electric power systems planning under uncertainty. The DSFP method can solve ratio
optimization problems involving issues of capacity expansion and
random information, where techniques of chance-constrained programming (CCP) and mixed integer linear programming (MILP) are
integrated into a linear fractional programming (LFP) framework.
An effective solution method has also been explored to tackle this
integrated model. The proposed DSFP approach has advantages in:
(a) balancing conicting objectives, (b) optimizing system efciency, (c) handling stochastic information expressed as probability distributions, and (d) generating exible capacity-expansion
strategies under different risk levels.
Applicability of the DSFP method has been demonstrated
through an expansion case study of municipal electric power generation system. The obtained solutions are effective in generating
sustainable power generation schemes and capacity expansion
plans with maximized system efciency under various constraint-violation risks. The results indicate that DSFP can incorporate valuable stochastic information into the process of decision
making, and generate dynamic management schemes with different levels of system reliability. Furthermore, it can support
in-depth analysis of the interactions among system efciency, economic cost and constraint-violation risk.

This study attempts to provide a modeling framework for solving ratio optimization problems involving issues of capacity expansions and random inputs. Although the proposed method is for the
rst time applied to the energy management led, it can be a powerful decision tool for other resources management problems. The
DSFP could be further enhanced through incorporating interval
analysis methods and fuzzy set theory into its framework.

Acknowledgements
This research was supported by the Program for Innovative Research Team (IRT1127), the MOE Key Project Program (311013),
and the Natural Science and Engineering Research Council of
Canada.

References
[1] Antunes CH, Martins AG, Brito IS. A multiple objective mixed integer linear
programming model for power generation expansion planning. Energy
2004;29:61327.
[2] Banzo M, Ramos A. Stochastic optimization model for electric power system
planning of offshore wind farms. IEEE Trans Power Syst 2011;26(3):133848.
[3] Bhavaraju MP, Hebson Jr JD, Wood W. Emerging issues in power system
planning. Proc IEEE 1989;77(6):8918.
[4] Cai YP, Huang GH, Yang ZF, Lin QG, Tan Q. Community-scale renewable energy
systems planning under uncertainty an interval chance-constrained
programming approach. Renew Sustain Energy Rev 2009;13:72135.
[5] Cai YP, Huang GH, Tan Q, Yang ZF. Planning of community-scale renewable
energy management systems in a mixed stochastic and fuzzy environment.
Renew Energy 2009;34(7):183347.
[6] Centeno E, Reneses J, Barqun J. Strategic analysis of electricity markets under
uncertainty: a conjectured-price-response approach. IEEE Trans Power Syst
2007;22(1):42332.
[7] Chang CT. A goal programming approach for fuzzy multiobjective fractional
programming problems. Int J Syst Sci 2009;40(8):86774.
[8] Charnes A, Cooper WW. Programming with linear fractional functionals. Nav
Res Logist Quart 1962;9:1816.
[9] Charnes A, Cooper WW, Kirby P. Chance constrained programming: an
extension of statistical method optimizing methods in statistics. New
York: Academic; 1972. p. 391402.
[10] Charnes A, Cooper WW, Rhodes E. Measuring the efciency of decision making
units. Eur J Oper Res 1978;2:42944.
[11] Cormio C, Dicorato M, Minoia A, Trovato M. A regional energy planning
methodology including renewable energy sources and environmental
constraints. Renew Sust Energy Rev 2003;7(2):99130.
[12] Diakoulaki D, Antunes CH, Martins AG. MCDA and energy planning. In:
Multiple criteria decision analysis: state of the art surveys. New
York: Springer; 2005. p. 85990.
[13] Fan YR, Huang GH. A robust two-step method for solving interval linear
programming problems within an environmental management context. J
Environ Inf 2012;19(1):19.
[14] Gardner DT. Flexibility in electric power planning: coping with demand
uncertainty. Energy 1996;21(12):120718.
[15] Gomez T, Hernandez M, Leon MA, Caballero R. A forest planning problem
solved via a linear fractional goal programming model. Forest Ecol Manage
2006;227:7988.
[16] Huang GH, Niu YT, Lin QG, Zhang XX, Yang YP. An interval-parameter chanceconstraint mixed-integer programming for energy systems planning under
uncertainty. Energy Sources Part B: Econ Plann Policy 2011;6(2):192205.
[17] Hladik M. Generalized linear fractional programming under interval
uncertainty. Eur J Oper Res 2010;205:426.
[18] Hsu GJY, Chou FY. Integrated planning for mitigating CO2 emissions in Taiwan:
a multi-objective programming approach. Energy Policy 2000;28:51923.
[19] Jeganathan C, Roy PS, Jha MN. Multi-objective spatial decision model for land
use planning in a tourism district of India. J Environ Inf 2011;17(1):1524.
[20] Karaki SH, Chaaban FB, Al-Nakhl N, Tarhini KA. Power generation expansion
planning with environmental consideration for Lebanon. Int J Electr Power
Energy Syst 2002;24:6119.
[21] Lara P, Stancu-Minasian IM. Fractional programming: a tool for the assessment
of sustainability. Agric Syst 1999;62:13141.
[22] Li GC, Huang GH, Lin QG, Zhang XD, Tan Q, Chen YM. Development of a GHGmitigation oriented inexact dynamic model for regional energy system
management. Energy 2011;36:338898.
[23] Li YP, Huang GH, Chen X. Planning regional energy system in association with
greenhouse
gas
mitigation
under
uncertainty.
Appl
Energy
2011;88(3):599611.
[24] Li YP, Huang GH, Sun W. Management of uncertain information for
environmental systems using a multistage fuzzy-stochastic programming
model with soft constraints. J Environ Inf 2011;18(1):2837.

H. Zhu, G.H. Huang / Electrical Power and Energy Systems 53 (2013) 553563
[25] Limmeechokchai B, Chungpaibulpatana S. Application of cool storage
airconditioning in the commercial sector: an integrated resource planning
approach for power capacity expansion planning and emission reduction. Appl
Energy 2001;68(3):289300.
[26] Lin QG, Huang GH. Planning of energy system management and GHG-emission
control in the municipality of Beijingan inexact-dynamic stochastic
programming model. Energy Policy 2009;37(11):446373.
[27] Liu L, Huang GH, Fuller GA, Chakma A, Guo HC. A dynamic optimization
approach for nonrenewable energy resources management under uncertainty.
J Petrol Sci Eng 2000;26(14):3019.
[28] Malik A. Modelling and economic analysis of DSM programs in generation
planning. Int J Electr Power Energy Syst 2001;23(5):4139.
[29] Marn A, Salmern J. A risk function for the stochastic modeling of electric
capacity expansion. Nav Res Log 2001;48:66283.
[30] Martins AG, Coelho D, Antunes CH, Climaco J. A multiple objective linear
programming approach to power generation planning with demand-side
management (DSM). Int Trans Oper Res 1996;3:30517.
[31] Mavrotas G, Diakoulaki D, Papayannakis L. An energy planning approach based
on mixed 01 multiple objective linear programming. Int Trans Oper Res
1999;6:23144.
[32] Mehra A, Chandra S, Bector CR. Acceptable optimality in linear fractional
programming with fuzzy coefcients. Fuzzy Optim Decis Making 2007;6:516.
[33] Nasiri F, Huang GH. Integrated capacity planning for electricity generation: a
fuzzy environmental policy analysis approach. Energy Sources Part B Econ
Plann Policy 2008;3(3):25979.
[34] Nilsson J, Mrtensson A. Municipal energy-planning and development of local
energy-systems. Appl Energy 2003;76(13):17987.
[35] Pereira AJ, Saraiva JT. Generation expansion planning (GEP) a long-term
approach using system dynamics and genetic algorithms (GAs). Energy
2011;36(8):518099.
[36] Ramanathan R, Ganesh LS. Energy alternatives for lighting in households: an
evaluation using an integrated goal programming AHP model. Energy
1995;20(1):6372.
[37] Romero C, Rehman T. Multiple criteria analysis for agricultural
decisions. Amsterdam: Elsevier; 1989.
[38] Rong A, Lahdelma R. An efcient linear programming model and optimization
algorithm for trigeneration. Appl Energy 2005;82(1):4063.

563

[39] Rong A, Lahdelma R. An effective heuristic for combined heat-and-power

production planning with power ramp constraints. Appl Energy
2007;84(3):30725.
[40] Sadeghi M, Hosseini HM. Energy supply planning in Iran by using fuzzy linear
programming approach (regarding uncertainties of investment costs). Energy
Policy 2006;34:9931003.
[41] Stancu-Minasian IM. Fractional programming: theory, methods and
applications. Dordrecht: Kluwer Academic Publishers; 1997.
[42] Stancu-Minasian IM. A fth bibliography of fractional programming.
Optimization 1999;5(14):34367.
[43] Tabors RD, Monroe BL. Planning for future uncertainties in electric power
generation: an analysis of transitional strategies for reduction of carbon and
sulfur emissions. Trans Power Syst 1991;6(4):15007.
[44] Unsihuay-Vila C, Marangon-Lima JW, Zambroni de Souza AC, Perez-Arriaga IJ.
Multistage expansion planning of generation and interconnections with
sustainable energy development criteria: a multiobjective model. Int J Electr
Power Energy Syst 2011;33:25870.
[45] Voropai N, Ivanova E. Multi-criteria decision analysis techniques in electric
power system expansion planning. Int J Electr Power Energy Syst
2002;24(1):718.
[46] Wang Q, Chen Y. Status and outlook of Chinas free-carbon electricity. Renew
Sustain Energy Rev 2010;14(3):101425.
[47] Zhou Q, Chan CW, Tontiwachiwuthikul P. Development of an intelligent
system for monitoring and diagnosis of the carbon dioxide capture process. J
Environ Inf 2011;18(2):7583.
[48] Zhu H, Huang GH. SLFP: a stochastic linear fractional programming approach
for sustainable waste management. Waste Manage 2011;31:26129.
[49] Zhu H, Huang GH, Guo P. SIFNP: simulation-based interval-fuzzy nonlinear
programming for seasonal planning of stream water quality management.
Water Air Soil Pollut 2012;23(5):205172.
[50] Zhu H, Huang GH, Guo P, Qin XS. A fuzzy robust nonlinear programming model
for stream water quality management. Water Resour Manage
2009;23:291340.
[51] Zhu Y, Huang GH, Li YP, He L, Zhang XX. An interval full-innite mixed-integer
programming method for planning municipal energy systems a case study
for Beijing. Appl Energy 2011;88:284662.