You are on page 1of 9

418 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 24, NO.

1, FEBRUARY 2009

A Stochastic Optimization Approach


to Rating of Energy Storage Systems
in Wind-Diesel Isolated Grids
Chad Abbey, Student Member, IEEE, and Géza Joós, Fellow, IEEE

Abstract—Wind-diesel systems represent a proactive step to- Power rating of energy storage system.
wards sustainable remote communities. However, for high ratios of
wind energy, the necessity of a dump load and the diesel operating Diesel minimum power constraint.
constraints need to be considered. Energy storage systems offer a
means of optimizing energy use and further reducing consumption Ratio of daily wind energy to load energy, or also
of diesel fuel. This paper proposes a methodology for storage sizing termed daily energy penetration.
based on stochastic optimization. The problem is formulated and
Wind installed power penetration level.
solved using representative data. The dependence of storage sizing
and the cost of delivered energy on wind penetration levels, storage Binary variable associated with diesel plant
efficiency, and diesel operating strategies are considered. Results dispatch at time, .
demonstrate that for high wind penetration, the availability of
storage, together with an appropriate diesel operating approach, Random variable associated with each scenario,
can result in significant cost savings in terms of fuel and operating relating the probability of a given combination of
costs.
, .
Index Terms—Energy storage, power generation planning,
power systems economics, stochastic systems, wind energy. Vector of first stage decision variables.
Vector of second stage decision variables.
I. NOMENCLATURE Efficiency of energy storage charging and
discharging processes.
The following describes the nomenclature used for various
variables and parameters throughout the paper. Boldface is used Price of energy supplied by diesel power, in
to denote vectors and matrices. $/kWh.
Cost of storage energy capacity, in $/kWh.
Energy state at time, .
Cost of storage power capacity, in $/kW.
Energy rating of energy storage system.
Price of energy supplied by wind power, in $/kWh.
Energy of the load over a 24-h period.
Hourly correlation coefficient between wind power
Initial energy state of energy storage system. and load.
Total delivered wind energy over a 24-h period.
Energy storage charging power at time, . II. INTRODUCTION

Diesel generator power at time, .


Dump load power at time, .
S USTAINED growth of wind suggests that it will ultimately
become a significant component of generation portfolios of
many of the major power systems in the near future. With this
expanding role, means to deal with its intermittency will need to
Energy storage system discharging power at time,
be invoked to ensure that the reliability of the system is upheld.
.
Isolated power systems are no exception, as even in remote com-
Load power at time, . munities people are turning to wind to help in absolving them-
Wind power at time, . selves of, or at least limiting their dependence on fossil fuels,
which is generally diesel in the case of remote power systems
[1]–[3].
Manuscript received March 12, 2008; revised July 11, 2008. First published This evolution has created a potential role for energy storage
December 16, 2008; current version published January 21, 2009. This work was technologies, with balancing power for intermittent renewables
supported in part by the Natural Sciences and Engineering Research Council of
Canada. Paper no. TPWRS-00200-2008.
energy touted as one of its possible applications. Although, in
The authors are with Power Engineering Research Laboratory, Department theory this is almost self-evident, the economic justification
of Electrical and Computer Engineering, McGill University, Montreal, QC H3A is not. These technologies remain onerous and therefore, they
2A7, Canada.
Color versions of one or more of the figures in this paper are available online
must be linked to a quantifiable value stream in order to gauge
at http://ieeexplore.ieee.org. the rating of the ESS that is required to ensure a sufficient
Digital Object Identifier 10.1109/TPWRS.2008.2004840 return. Also, given that the energy storage system (ESS) must
0885-8950/$25.00 © 2008 IEEE
ABBEY AND JOÓS: A STOCHASTIC OPTIMIZATION APPROACH TO RATING OF ENERGY STORAGE SYSTEMS IN WIND-DIESEL ISOLATED GRIDS 419

be rated both in terms of power and energy capability, and


that wind energy is a stochastic energy source, increases the
dimensions of the problem.
Energy storage system sizing has been approached in much of
the literature by first considering optimal scheduling of the device
and then performing sensitivity analyses to determine the impact
of different ratings [4]–[10]. The parametric studies either inves-
tigate the impact on reliability indices [4] or an objective function
[5]–[8], which usually relates to the profit realized by the instal-
lation, or which may be a metric related to the performance of the
power system [9]. The results of these studies are quite useful
and can serve as an approach to optimal scheduling of the ESS.
The problem of storage sizing for remote communities has
been considered in the context of design of the energy supply
Fig. 1. Remote wind-diesel power system with an energy storage system.
mix of the community to minimize the cost of energy served
[11]–[13]. In each of these references, both operational costs
and fixed costs are included in the optimization problem for-
mulation. Databases of monthly energy production from the dump load is used to manage the wind provided to the commu-
renewables energy sources (wind and photovoltaics) are used nity, or stated otherwise, to regulate the loading on the diesel
to estimate the contribution from these sources. Reliability de- unit(s).
sign criteria are then imposed through the use of adequacy con- This will generally remain unchanged in the presence of ESS,
straints, which can be met through the inclusion of storage or even though the balance of plant could be shared with the ESS or
additional diesel capacity. Although useful as a pre-feasibility even accorded solely to the ESS for limited periods. In this paper
assessment for energy storage, each approach neglects intraday we will consider the results for both status quo (diesel is always
operational issues. Also, the different operating schemes of the scheduled on) as well as unconventional operating schemes.
diesel back-up unit are not dealt with exhaustively, leaving some 2) Wind and Load Characteristics: The problem of energy
unanswered questions. storage sizing is related to tapping into a revenue stream that
An alternative is the use of stochastic optimization tech- allows the developer to recover the costs associated with the in-
niques. These have been applied to various problems in power stallation (as well as operational costs). For remote power sys-
systems to address uncertainties. In [14], Galiana et al. pre- tems this revenue is achieved by avoiding the dumping of energy
sented both the deterministic and stochastic formulations for via the dump load. The degree of dumped wind energy depends
the problem of scheduling different power system reserves. on: the diesel operating strategy (and how this impacts the op-
The scheduling of a hydro plant in a thermal based system was erational constraints of the wind park), the amount of wind en-
solved using a stochastic formulation of the problem in [15], ergy produced, and its correlation with respect to the community
while in [16], the dispatch of distributed generation units was load.
considered using the same theory. An overriding assumption is that the most important cycling
This paper considers the problem of sizing of ESS for iso- of the ESS occurs over daily periods and as a result the rela-
lated wind-diesel power systems using stochastic optimization. tionship between load and the wind resource needs to be estab-
In Section III, the problem is described and formulated as a sto- lished over this interval. To facilitate the modeling of the wind
chastic optimization problem to capture the uncertainty associ- resource and its relationship to the community load, we will re-
ated with the daily energy penetration of the wind and its corre- quire a number of metrics. The wind power penetration will be
lation with the load. The problem is posed firstly with only con- defined as
tinuous variables, and then as a mixed-integer formulation that
considers unit commitment of the diesel plant. In Section IV, re- (1)
sults are given along with a sensitivity analysis to consider the
impact of wind power penetration, ESS efficiency, and opera- where is the installed capacity of the wind park and is
tional modes on the sizing of the ESS and the cost of the energy the peak load of the community. The daily energy penetration,
served. which provides a measure of the amount of wind energy pro-
duced during a given day compared to the total load for the day
III. PROBLEM DESCRIPTION will be given by

A. System Overview (2)


1) Wind-Diesel System: The majority of remote systems con-
sist of a distributed load that is supplied by diesel generators, Finally, the correlation of the two profiles is vital as it pro-
which serves as the balance of plant. Other elements of the re- vides an indication of when the peak of the wind power occurs
mote power system may include: renewable energy (wind, pho- relative to the load peak, given by
tovoltaics, small hydro), a dump load possibly one or more ESS
technologies (see Fig. 1). The diesel plant remains the control (3)
element for frequency and voltage for wind-diesel systems. The
420 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 24, NO. 1, FEBRUARY 2009

Fig. 2. Methodology for energy storage system sizing for integration of wind.
Fig. 3. Scenario tree for stochastic optimization approach to ESS sizing.

where the subscripts and refer to the hourly wind power and
load time series; and denote the average value of the wind 1) Scenario Generation: As mentioned above, each scenario
and load; and correspond to their standard deviations, is a 24-h period of wind and load for the community. Each by
and . itself constitutes a deterministic solution of the problem. In the
3) Diesel Operating Strategies: The operation strategy of the two-stage stochastic formulation [17], the second stage vari-
diesel generator will play an important role in determining the ables are specified for each scenario, while the first stage vari-
magnitude of dumped wind energy and consequently, the sizing ables are common to each. For our problem, the scheduling
of the energy storage. Three different diesel operating strategies for each scenario is flexible—one is not obliged to consider
will be considered, namely, minimum loading; low-load diesel how storage was scheduled the month before1 —whereas rat-
technologies; and diesel unit commitment. ings must be defined before the installation is constructed, and
Perhaps the most widespread is the use of a dump load in obviously is shared across scenarios (see Fig. 3).
order to ensure that the diesel generator is loaded at or above a We define the random variable associated with each
minimum value (30% of its rated capacity is typical). Generally of the scenarios, which is the probability of a wind and load
operation below these levels is avoided as it leads to reduction daily profile having a certain correlation coefficient, and of
in lifetime, may lead to fire hazards, and the efficiency gener- a certain energy ratio, . Assuming that and are
ally degrades at low loadings [2]. However, remote communities independent the probability of is given by
often have large differences between their peak and minimum
loads; even in the absence of wind, diesels may be forced below
(4)
this threshold.
New diesel generator technologies that utilize electronic fuel
injectors can lead to much improved operation well below 0.3 whose distribution can be plotted through analysis of databases
per unit, as they are able to maintain sufficient engine temper- of wind power and load data. Obviously the larger the database
ature to avoid the build-up of particulate matter and can selec- the better; however it should at least span an entire year to ac-
tively reduce the number of pistons in operation. These units count for seasonal variations of the wind resource and the load.
will be modeled by simply relaxing the low loading constraint. 2) ESS Storage Financing Model: Parameters for storage
Diesel unit commitment refers to the case where there is suffi- characteristics and energy prices are required as part of the eco-
cient wind power and stored energy (and power capacity) to shut nomic calculations. In Section IV-A, a design example is uti-
the diesel off during specific intervals. This is modeled using the lized to illustrate the concepts for realistic variables. Using the
binary variable, , which will be designated “1” when the fixed costs from Table II, the daily incremental costs associated
diesel is in operation and “0” when it is shutdown. As will be with amortization of the investment were calculated using the
seen in the next section, these three modes imply different for- following rudimentary calculation of annuity:
mulations for the optimization problem.
(5)
B. Methodology
A generalized representation of the proposed methodology is which converts to a $/kWh/day amount. The project
given in Fig. 2. The wind and load resource must be character- period, , is 20 years and the interest is compounded annually,
ized in terms of random values and the optimization problem using a discount rate of 8.5%. The same approach was applied
is solved to determine what is the sizing of and . This for the power rating. These parameters are considered further in
analysis considers scenarios of daily load and wind power pro- the formulation of the optimization problem.
files, weighted with the probabilities of their occurrence. Gen- 1There is actually a link between different days through the stored energy in
eration of these probabilities is described in the following sub- the ESS at the end of the day, e . This is accounted for in the formulation by
section. defining a third first stage variable E and a constraint that links it with e , (17).
ABBEY AND JOÓS: A STOCHASTIC OPTIMIZATION APPROACH TO RATING OF ENERGY STORAGE SYSTEMS IN WIND-DIESEL ISOLATED GRIDS 421

TABLE I The vector of first stage variables, , is given by


ENERGY PRICE DATA AND DIESEL CONSTRAINTS FOR BASE CASE
(7)

The rating of the ESS in terms of its power and energy, and
the initial energy state of the storage device, . The second
stage variables are given by the vector,
TABLE II
ENERGY STORAGE SYSTEM DATA FOR BASE CASE (8)

Representing the diesel, dump load, and ESS charging and


discharging powers for each time interval in the scenario, with
probability .
The optimization problem is subject to the following con-
*Typical ESS parameters were taken from [21] and corroborated using [22]. straints.
1) Power Balance:
C. Problem Formulation
The sizing of the ESS must rationalize the costs associated
with the installation (both fixed and operating costs) with the (9)
monetary benefit that it brings to the system. In many cases there
are less tangible benefits that although real are not monetized Typical distribution losses are small (5%) but may be ex-
and attributed to the ESS. However, we will concern ourselves treme in some cases (15%). However, the effect in terms of the
with those where a real value stream is in place. The problem is problem is simply that of additional load. There will be a small
to determine how much storage capacity is required in order to effect on storage sizing (slightly smaller rating) but it has been
minimize the supplied daily energy of a wind-diesel plant. neglected here.
Before proceeding further we state the following assump- 2) Diesel Constraints:
tions:
(10)
• and are uncorrelated;
• dumped energy is lost as waste heat, i.e., it is not used to Or in the case where the diesel is allowed to shutdown—this
source local heating loads; assumes that the balance of plant functionality is performed by
• direct load control is not considered. the ESS during these periods—then (10) becomes
The first is required only to use the simplified equation for
calculation of the , if it is not true it only complicates the (11)
calculation of these probabilities. The second makes the ESS
sizing case optimistic as the revenue for avoided dump load is This introduces binary variables, which complicates the
equivalent to the cost of producing that energy. Systems do exist solution of the optimization problem, requiring mixed integer
where this is not valid and the formulation would need to be solvers, such as [18].
modified in consequence. Demand-side management (DSM) or 3) Dump Load Constraints:
direct load control (DLC) could also be employed as alternatives
to or in concert with ESS; however, they are excluded from the (12)
current formulation.
Bearing these assumptions in mind, the problem can be stated 4) ESS Power Constraints:
formally as
(13)

(14)
(6)
5) ESS Energy Rating Limits:

where the operating costs are calculated over all time points, (15)
. The operator calculates the expected value of the
operating costs over the random variable, . The cost associated 6) Energy Transition:
with wind power is included in the objective function. However,
the cost of wind energy has no bearing on the rating of the energy (16)
storage system due to the fact that the decision variables can in
no way impact the quantity of wind energy produced. The cost
of wind energy is only included to calculate a representative cost Also, as a means of respecting the fact that the storage should
of energy served. not completely neglect the requirements of the following day,
422 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 24, NO. 1, FEBRUARY 2009

a constraint is imposed on the energy at the end of the profile,


such that it must be equivalent to the initial energy

(17)

This final constraint, although required here in the ESS rating


problem need not necessary be respected by an online oper-
ating algorithm. For the problem under discussion, this con-
straint serves to link the different scenarios using the energy in
the device at the end (or beginning) of the interval. Another point
worth mentioning is that neither nor appear in the objec-
tive function. A modification to the formulation could include
these if it were of interest to penalize certain states-of-charge,
for instance deep discharges. This would require more in-depth Fig. 4. Discrete probability distribution of the correlation factor between daily
wind power and load.
knowledge of the characteristics of the ESS technology in ques-
tion and appropriate costs associated with these specific oper-
ating points.

IV. RESULTS AND DISCUSSION


The formulation of the problem is now applied to a specific
case to illustrate the application of the model and the advantage
of a stochastic formulation of the problem. This is followed by
parametric analysis where the impact of key parameters on the
solution are considered, namely: installed wind power capacity,
ESS efficiency, and the operating strategy of the diesel plant.

A. Design Example
A design example is utilized for the investigation and to
permit verification of models in the event the model is to be Fig. 5. Discrete probability distribution of daily wind energy penetration, r
reproduced by others. The data requirements are discussed in for r = 05: .
the following subsections.
1) Economic Parameters: Input parameters for the economic
models are included in Tables I and II. The energy prices as- correlation gives a general idea of how the two are matched,
sociated with diesel fuel and small wind vary significantly in their daily correlation will dictate the magnitude of the dumped
remote communities but the values chosen fall in the set of re- energy.
alistic values, as supported by [19] and [20], respectively. After The distributions of and were then calculated (see
establishing a realistic range, the ultimate values for the base Figs. 4 and 5). These were divided into discrete divisions, the
case were selected such that the sizing was favorable (as op- number depending on the total number of scenarios to be mod-
posed to results where the ESS sizing was zero). This was done eled. Obviously, the greater number of scenarios used the closer
to facilitate the analysis but was deemed reasonable as the cost the results will represent the true data, the tradeoff being com-
of diesel is expected to rise whereas the cost of wind energy will putational efficiency. The actual wind profile used for each sce-
also rise but at a more moderate rate. As the cost of diesel was nario was generated by multiplying a base load profile by
varied, the ratio of the cost of wind energy to diesel energy was and then time shifting to achieve the desired . It is important
kept constant at two-thirds. to note that the resulting wind profile is not unique, as there are
2) Wind and Load Characteristics: Ideally the wind and load an infinite number of wind profiles that could be used to obtain
data should be collected over several years, but for practical rea- the required and .
sons (e.g., budget, time constraints) this may not be possible. As
a minimum, one year of data should be used in order to capture B. Base Case Results
seasonal variations. The wind and load data for the current study A rudimentary approach to sizing is to solve a deterministic
was taken from references [23] and [24], respectively. These formulation of the problem considering the expected values of
sources provide hourly wind speeds and load magnitudes for all and . This approach could be extended to all 365 days.
hours of the year. However, this is equivalent in computational intensity to 365
The wind speeds were converted to powers using the wind scenarios of the stochastic formulation, and if multiple years are
power curve from the Entergrity 50-kW wind turbines [25]. The included the problem soon becomes intractable. Therefore, the
wind park model applies a scaling factor for the number of tur- stochastic solution captures the probabilities of these two vari-
bines, neglecting array losses and forced outage rates. The wind ables, achieving similar levels of accuracy but with significantly
power tends to peak in the winter months whereas the load has less computational requirements than simulation of the full data
peaks in both the winter and summer months. While the seasonal set.
ABBEY AND JOÓS: A STOCHASTIC OPTIMIZATION APPROACH TO RATING OF ENERGY STORAGE SYSTEMS IN WIND-DIESEL ISOLATED GRIDS 423

TABLE III TABLE IV


ENERGY STORAGE SYSTEM SIZING BASE CASE RESULTS DEPENDENCE OF COST OF ENERGY SERVED AND ESS SIZING ON
CONSIDERING THE NUMBER OF SCENARIOS CHANGES IN CAPITAL COSTS, DIESEL ENERGY COSTS, AND HOURLY
CORRELATION BETWEEN WIND AND LOAD

*When not indicated, the values of the parameters are as indicated for the base
case.

and an increase in diesel fuel price are considered, which rep-


As the number of cases is increased, the formulation more resent possible future scenarios. In all cases, this leads to an in-
accurately represents the combined probability distributions of crease in the ESS ratings. Reductions in ESS capital costs have
the two random variables, and . The number of sce- a positive benefit on the price of energy served, whereas higher
narios becomes sufficient when the ESS ratings converge and diesel costs result in higher energy costs as it still represents the
the value of additional scenarios—given by the difference be- largest share of the generated energy.
tween expected costs and true costs (calculated using the next The extremes cases for wind-load correlation were consid-
level of scenarios)—is sufficiently small. The more scenarios, ered. As load and wind are perfectly matched, the ESS sizing
the greater the computational burden and thus a tradeoff must drops to 0, as the dump load is used sparingly, effectively elim-
be made at some point. In contrast, a Monte Carlo approach inating the value stream. Negatively correlated patterns imply
would use thousands of simulations, each coming up with op- greater use of the dump load, and consequently a higher cost of
timal values for the ESS ratings. This would yield a distribution energy but also greater ESS capacity. The relationship between
of energy storage ratings, with the expected value being the op- load and wind is critical in identifying sites where the conditions
timal rating of the device. for ESS are favorable.
Table III gives the results for the base case, for increasing 2) Installed Wind Capacity: The results associated with in-
number of scenarios, for both the continuous and binary formu- creasing wind penetration are revealing in that they show that
lations of the problem. All formulations of the problem were de- in certain cases ESS sizing is nonzero, whereas as at low pen-
fined in MATLAB and solved using an interface to GAMS, [18]. etrations, it is not at all justified. The most interesting range is
There are a couple of important observations to make. First, the medium to high penetration levels, as can be seen in the en-
the ratings of and both decrease with the number of ergy cost figures and rating, Figs. 6 and 7, respectively. There is
scenarios and eventually level off once four or more slices are an optimum penetration level (for the present case it occurs at
used for each distribution (25 scenarios). Also, the true cost of ) where ESS brings the greatest benefit and conse-
energy2 decreases in a similar manner as the ratings. This pro- quently the sizing is the largest.
vides a means of assessing when additional scenarios no longer However, one must bear in mind that the overall benefit of
provide additional benefit. A third observation is that there is a reduction of the cost of energy is very marginal—it is dubious
significant difference in cost for the binary and continuous so- that the project would go forward given the associated risks (site
lutions, a fact that will be explored in greater detail later. access issues, new technology). One can hope that reduction in
capital costs would help make things more attractive. The case
C. Sensitivity Analysis where capital costs are neglected in face gives the maximum
In this section the impact of economic parameters, the wind reduction in $/kWh possible. The sizing in this case is quite dif-
resource, ESS efficiency, and diesel operating strategy are con- ferent as the concern becomes only optimizing the use of energy,
sidered. The problem is solved using the base case values and for with the peak sizing occurring when the real sizing should ac-
two additional cases when relevant: the case with no ESS and tually be zero.
when the capital costs of the ESS are neglected.3 These latter Interestingly, at extremely high penetration levels, with in-
two cases represent extremes and facilitate interpretation of the stalled wind exceeding peak load, the ESS sizing again becomes
results and discussion. nonzero, with an increased ratio of energy sizing to power. The
1) Economic Parameters and Wind Load Correlation: In storage attempts to shift all amounts of wind energy so that the
Table IV, the effect of a reduction in capital costs of the ESS diesel always at its minimum loading (note that in Fig. 8, diesel
energy approaches 0.3). This is due to the fact that the cost of
2This is calculated by fixing the first stage variables to the determined optimal
wind and diesel power are not equal—were it the case, the sizing
for that case and running again using the maximum number of scenarios (81
scenarios). would remain zero. However, these wind penetration levels are
3A nominal value (1/10000  ) was used to limit ESS sizing this case. Ef- difficult to justify as the costs again approach the cost of diesel
ficiency losses also serve to restrict ESS sizing. (0.60$/kWh).
424 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 24, NO. 1, FEBRUARY 2009

Fig. 6. Cost versus wind penetration for: no ESS (dotted line), ESS (base case),
Fig. 9. Cost versus wind penetration for: no ESS (dotted line), ESS base case
and ignoring the capital cost of ESS (heavy dashed line).
(solid line), and considering diesel generator shutdown (heavy dotted line).

}
Fig. 7. Energy storage power (“O”) and energy (“ ”) ratings versus wind pen-
}
Fig. 10. Energy storage power (“O”) and energy (“ ”) ratings versus wind
penetration for base case (solid line) and considering diesel generator shutdown
etration for base case (solid line) and ignoring capital costs (dotted line).
(dotted line).

changes are small. For the base case values, storage is not jus-
tified for efficiencies below 0.65 while improvements to
result in a less than 1 percent reduction in cost. Beyond 0.85,
the overall rating of the base case changes only marginally, sug-
gesting a design requirement in efficiency may only be to meet
a lower limit on .
4) Diesel Operating Strategy: Similar cases as in
Section IV-C.II were repeated, only now considering the
additional benefit that can be realized by allowing the ESS
to replace the diesel as the balance of plant during certain
intervals. The results are almost identical up to the point where
storage first becomes justified, beyond which the costs are
Fig. 8. Wind energy, diesel energy generation, total generated energy, dumped greatly reduced—at the cost of energy supplied is
energy, and storage energy rating versus wind penetration. reduced by nearly 12% (see Fig. 9). Furthermore the optimal
penetration level shifts further to the right ( com-
pared with 0.9) facilitating more efficient operation at extreme
Fig. 8 shows the main energy components as a function penetration levels.
of wind penetration. It shows that ESS is only justified after Considering the ratings of the ESS (see Fig. 10), it can be
dumped energy becomes sufficiently large. Use of diesel en- noted that the energy ratings for the case with binary variables
ergy decreases, due to increased wind energy and the ability of at lower penetration levels are actually lower, whereas the peak
storage to limit the dumped energy. At extreme penetrations rating for the binary case occurs when the rating for the con-
diesel energy approaches 0.3 as stated above. tinuous case returns to zero. Plots of energies (see Fig. 11) re-
3) Storage Efficiency: Considering the case of , inforce the observations from these results. The total generated
the role of storage efficiency was then considered; the results energy actually remains constant (dump load remains zero) up
are summarized here. In general, the cost of energy reduces and to . This is reflected in the steep drop in diesel en-
ESS becomes more viable as efficiency improves, however, the ergy used, which is 20% lower at the upper end.
ABBEY AND JOÓS: A STOCHASTIC OPTIMIZATION APPROACH TO RATING OF ENERGY STORAGE SYSTEMS IN WIND-DIESEL ISOLATED GRIDS 425

be established; this value being 0.85 for the base case consid-
ered. Reduction of ESS capital cost appears to more importantly
translate to reduced energy costs. Perhaps the most striking re-
sults are the differences associated with various diesel operating
schemes, particularly when the diesel is allowed to shut-down
and let the ESS serve as the balance of plant. Under this case
the cost reduction compared with the base case is roughly 12%.
Taken together with the reduction of dumped energy, these re-
sults suggest that the combination of ESS and innovative diesel
operating practices can lead to not only a marketable reduction
in cost but also to more sustainable approaches of serving the
energy needs of remote communities.
Fig. 11. Wind energy, diesel energy generation, total generated energy, dumped
energy, and storage energy rating versus wind penetration considering diesel ACKNOWLEDGMENT
generator shutdown.
The authors would like to thank J. Restrepo for his valuable
comments and help with the use of the optimization problem
solver.

REFERENCES
[1] R. Hunter and G. Elliot, Wind-Diesel Systems: A Guide to the Tech-
nology and Its Implementation. Cambridge, U.K.: Cambridge Univ.
Press, 1994.
[2] F. Katiraei and C. Abbey, “Diesel plant sizing and performance anal-
ysis of a remote wind-diesel microgrid,” in Proc. IEEE Power Eng. Soc.
General Meeting, Jun. 24–28, 2007.
[3] P. Lautier, M. Prévost, and P. Martel, “Off-grid diesel power plant ef-
ficiency optimization and integration of renewable energy sources,” in
Proc. IEEE Can. Electrical Power Conf., Montréal, QC, Canada, Oct.
2007.
[4] Bagen and R. Billinton, “Incorporating well-being considerations in
generating systems using energy storage,” IEEE Trans. Energy Con-
vers., vol. 20, no. 1, pp. 225–230, Mar. 2005.
Fig. 12. Cost versus diesel minimum loading constraint for base case (dotted [5] E. D. Castronuovo and J. A. P. Lopes, “On the optimization of the daily
line) and considering diesel generator shutdown (solid line). operation of a wind-hydro power plant,” IEEE Trans. Power Syst., vol.
19, no. 3, pp. 1599–1606, Aug. 2004.
[6] N. Lu, J. Chow, and A. Desrochers, “Pumped-storage hydro-turbine
As relinquishing control of the power system’s performance bidding strategies in a competitive electricity market,” IEEE Trans.
Power Syst., vol. 19, no. 2, pp. 834–841, May 2004.
to the ESS is still a fairly liberal move, the impact of using ad- [7] M. Korpas and A. Holen, “Operation planning of hydrogen storage con-
vanced low-load diesel was also considered. The costs for the nected to wind power operating in a power market,” IEEE Trans. En-
base case and diesel unit commitment modes initially start at the ergy Convers., vol. 21, no. 3, pp. 742–749, Sep. 2006.
[8] M. Korpaas, A. Holen, and R. Hildrum, “Operation and sizing of energy
same point but as increases the two diverge (see Fig. 12). storage for wind power plants in a market system,” Elect. Power Energy
The base case increases almost linearly to the value obtained Syst., vol. 25, pp. 599–606, 2003.
in the first section. Costs in the unit commitment case remain [9] J. P. Barton and D. G. Infield, “Energy storage and its use with inter-
mittent renewable energy,” IEEE Trans. Energy Convers., vol. 19, no.
nearly constant. 2, pp. 441–448, Jun. 2004.
[10] M. Muselli, G. Notton, and A. Louche, “Design of hybrid-photovoltaic
V. CONCLUSIONS power generator, with optimization of energy management,” Sol. En-
ergy, vol. 65, no. 3, p. 143, 1999.
This paper has considered the problem of energy storage [11] R. S. Garcia and D. Weisser, “A wind-diesel system with hydrogen
system sizing for isolated wind-diesel power systems. The storage: Joint optimisation of design and dispatch,” Renew. Energy, vol.
31, no. 14, pp. 2296–2320, 2006.
problem has been formulated as a two-stage stochastic opti- [12] E. Koutroulis, D. Kolokotsa, A. Potrirakis, and K. Kalaitzakis,
mization problem, with the objective of minimizing the cost of “Methodology for optimal sizing of stand-alone photovoltaic/wind-
generator systems using genetic algorithms,” Sol. Energy, vol. 80, no.
supplied energy. The approach has been applied to a represen- 9, pp. 1072–1088, 2006.
tative case study and the impact of various parameters on ESS [13] A. Prasad and E. Natarajan, “Optimization of integrated photovoltaic-
sizing and energy costs were investigated, including: wind pene- wind power generation systems with battery storage,” Energy, vol. 31,
no. 12, pp. 1607–1618, 2006.
tration, ESS efficiency, and diesel operating strategy. For the base [14] F. Galiana, F. Bouffard, J. Arroyo, and J. Restrepo, “Scheduling
case, the stochastic formulation of the problem resulted in a two- and pricing of coupled energy and primary, secondary, and tertiary
thirds reduction in ESS power rating and half the energy rating reserves,” Proc. IEEE, vol. 93, no. 11, pp. 1970–1983, Nov. 2005.
[15] M. P. Nowak and W. Romisch, “Stochastic Lagragian relaxation
compared with the deterministic case (using the expected values). applied to power scheduling in a hydro-thermal system under uncer-
Sensitivity analysis revealed that ESS has a role to play in tainty,” Ann. Oper. Res., vol. 100, no. 1, pp. 251–272, Dec. 2000.
[16] V. Pappala and I. Erlich, “Management of distributed generation units
medium to high penetration scenarios but that the expected re- under stochastic load demands using particle swarm optimization,” in
duction in the cost of energy served (less than 1%) would not Proc. IEEE Power Eng. Soc. General Meeting, Tampa Bay, FL, Jun.
normally be sufficient to justify inclusion of ESS in the design. 24–28, 2007.
[17] J. Birge and F. Louveaux, Introduction to Stochastic Programming.
Storage efficiency does not significantly influence the cost; how- New York: Springer, 1997.
ever, results show that a lower bound on one way efficiency can [18] Gams Website, 2008. [Online]. Available: http://www.gams.com/.
426 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 24, NO. 1, FEBRUARY 2009

[19] M. Devine, E. Baring-Gould, and B. Petrie, “Wind-diesel hybrid Géza Joós (M’82–SM’89–F’06) received the M.Eng.
options for remote villages in Alaska,” in Proc. AWEA Annu. Conf., and Ph.D. degrees from McGill University, Montréal,
Chicago, IL, 2004. QC, Canada.
[20] Survey of the Small Wind (300 W to 300 kW) Turbine Market in He has been a Professor with McGill Univer-
Canada, 2008. [Online]. Available: http://www.smallwindenergy.ca. sity since 2001. He is involved in fundamental
[21] L. Mears and H. Gotschall, EPRI-DOE Handbook of Energy Storage and applied research related to the application
for Transmission and Distribution Applications. Washington, DC: of high-power electronics to power conversion,
EPRI and the U.S. DOE, 2003. including distributed generation and power sys-
[22] Electricity Storage Association Website, 2008. [Online]. Available: tems. He has published extensively and presented
http://electricitystorage.org. numerous papers and tutorials on these topics. His
[23] The Kansas Electric Utilities Wind Program, Kansas Electric Utilities employment experience also includes ABB, the
Research Program (KERUP), Wind Energy in Kansas, 2008. [Online].
University of Quebec, and Concordia University. He has been involved in
Available: http://www.agroplastics.com/kswind.
[24] C. Grigg et al., “The IEEE reliability test system-1996. A report pre- consulting activities in power electronics and power systems, and with CEA
pared by the reliability test system task force of the application of prob- Technologies as the Technology Coordinator of the Power Systems Planning
ability methods subcommittee,” IEEE Trans. Power Syst., vol. 14, no. and Operations Interest Group.
3, pp. 1010–1020, Aug. 1999. Dr. Joós is active in a number of IEEE Industry Applications Society com-
[25] Entegrity Wind Systems, Inc., EW15 Specifications, 2008. [Online]. mittees and in the IEEE Power Engineering Society and CIGRE activities and
Available: http://www.entegritywind.com. working groups dealing with power electronics and applications to distributed
resources. He is a Fellow of the Canadian Academy of Engineering.

Chad Abbey (S’01) received the degree B.Sc.


degree in electrical engineering from the University
of Alberta, Edmonton, AB, Canada, in 2002 and the
M.Eng. degree from McGill University, Montréal,
QC, Canada, in 2004, where he is currently pursuing
the Ph.D. degree.
His current research interests include wind energy,
distributed generation, energy storage, and their inte-
gration to the grid. He is presently working with the
CANMET Energy Technology Centre, in Varennes,
QC, where he is a Research Engineer and helps to co-
ordinate a joint research program on the modeling and integration of distributed
generation.
Mr. Abbey is an active member of CIGRE.

You might also like