Professional Documents
Culture Documents
2011 Economic Feasibility and Optimisation of An Energy Storage System For Portland Wind Farm (Victoria, Australia)
2011 Economic Feasibility and Optimisation of An Energy Storage System For Portland Wind Farm (Victoria, Australia)
Applied Energy
journal homepage: www.elsevier.com/locate/apenergy
a r t i c l e i n f o a b s t r a c t
Article history: This paper presents the details of a theoretical study of the economic advantages of using large-scale
Received 5 October 2010 energy storage to complement a wind farm in a base-load dominated electricity grid. A computer model
Received in revised form 25 November 2010 is developed which simulates the operation of several energy storage systems when used with the 190-
Accepted 3 December 2010
MW Portland Wind Farm (PWF) located in Portland, Victoria, Australia. A variety of operating strategies
Available online 19 February 2011
are compared with the results of a dynamic programming model which finds the maximum possible rev-
enue which a given system can generate for a set of input conditions. Three energy storage systems are
Keywords:
modelled and costed: Pumped Seawater Hydro Storage (PSHS), Compressed Air Energy Storage (CAES),
Wind power
Energy storage
and Thermal Energy Storage (TES). It is found that CAES is the most profitable storage medium, requiring
Economic feasibility a capital expenditure of A$140 M and generating a rate of return (ROR) of 15.4%. The ROR for PSHS was
Spot price analysis 9.6%, and for TES was 8.0%. Therefore, a significant investment opportunity exists for the installation of an
energy storage system in this wind farm. It is therefore highly recommended that CAES is investigated
further with the aim of introducing large-scale energy storage to PWF and other similar wind turbine
installations.
Crown Copyright Ó 2010 Published by Elsevier Ltd. All rights reserved.
1. Introduction 18.478 (Australia), 18.376 (USA) and 16.530 (Canada) and 21.269
(Luxembourg). Australia’s Federal Government has acknowledged
It is common knowledge among most scientists and engineers this ranking and has invested in renewable energy to reduce the
that humans are living unsustainably. Governments and industry country’s greenhouse gas emissions.
worldwide have acknowledged that the increasing carbon dioxide The vast growth in demand for electricity provides a significant
levels in the atmosphere are caused by human activity, and that opportunity for engineers and scientists to acquire more power
the constantly increasing magnitude of this emission will trigger from renewable sources. Growth in renewable energy is being en-
significant changes in the Earth’s climate. The consumption of hanced by recent government initiatives. The European Union has
non-renewable fuels to provide energy is a major source of carbon a target of 22% of total electricity production from renewable en-
dioxide and other greenhouse gases which cause global warming ergy sources by 2010 [6]. California has increased its standard for
[1,2]. Electricity comprises a large portion of the energy utilised renewable resources from 20% by 2010 to 33% by 2020 [7]. The
by humanity, estimated at 12% in 2008, and expected to grow to Australian Federal Government has, very disappointingly, lagged
34% by 2025 [3]. In Australia, the largest increase in CO2 emissions behind these initiatives. However, in 2001 the Mandatory Renew-
over the 1990–2007 period occurred in the stationary energy sec- able Energy Target (MRET) scheme was introduced, and in 2007
tor (an increase of almost 50%), and in 2007, stationary energy ac- the Federal Government committed to ensuring that 20% of Austra-
counted for 53.9% of national emissions [4]. As reported in [5], lia’s electricity supply comes from renewable energy sources by
Australia, among the OECD countries, is the largest emitter of 2020. Legislation was passed in October 2009 to implement an ex-
greenhouse gas per capita followed by USA and Canada with the panded National Renewable Energy Target, which should further
exclusion of Luxembourg which is considered to be reasonable as stimulate growth in renewable energies.
its total emission has an insignificant impact on the global green- Wind power is currently one of the most popular and fast-
house gas inventories. The per capita greenhouse gas emissions expanding sources of renewable energy. Current projections put
in tonnes of CO2 equivalent for these countries in 2008 were the average global growth of the industry at 24% pa for the next
5 years [8]. With large expanses of coastline available (character-
⇑ Corresponding author. Tel.: +61 3 9905 3562; fax: +61 3 9905 1825. ised by reliably high speed, low turbulence winds), Australia has
E-mail address: akbar.hessami@monash.edu (M.-A. Hessami). excellent wind resources by world standards, and is in a good posi-
0306-2619/$ - see front matter Crown Copyright Ó 2010 Published by Elsevier Ltd. All rights reserved.
doi:10.1016/j.apenergy.2010.12.013
2756 M.-A. Hessami, D.R. Bowly / Applied Energy 88 (2011) 2755–2763
tion to expand its use of wind power [4]. Wind power is a truly providing power to the grid during the peak demand periods when
renewable resource, producing no greenhouse emissions during electricity spot price is high.
its generation. Therefore, this study investigates the economic viability of
Like all energy sources, however, wind power has strengths and using an energy storage system with Portland Wind Farms (PWF)
weaknesses. One of the biggest drawbacks is that supply of wind is in Victoria, Australia. This wind farm is considered to be Victoria’s
inherently variable as it cannot provide ‘‘firm power’’ on its own largest wind farm by nameplate capacity (194.4 MW) when com-
[9]. Essentially, while wind can be reliably forecast several days pleted in 2010 [12]. For this purpose, three methods of energy stor-
in advance, it cannot be scheduled to meet grid demand. It is clas- age are investigated and compared: Pumped Seawater Hydro
sified as ‘‘negative load’’ (or ‘‘non-dispatchable’’) by electricity reg- (PSHS), Compressed Air Energy Storage (CAES), and Sensible Ther-
ulators, not to be relied upon to meet demand at critical times. mal Energy Storage (TES). These technologies are all commercially
In electricity grids with only small amounts of wind power gen- proven, although not all have previously been used on this scale.
eration, where the variability of wind cannot be ‘‘smoothed’’ by the A large body of research in the use of energy storage to comple-
use of geographically separated wind farm locations, another gen- ment wind power has been undertaken and reported in the litera-
eration source must be available in the grid to provide power in the ture. Many studies have investigated technical and power control
absence of wind. In the Victorian power grid, such sources take the aspects, and a variety of papers have presented cases for using en-
form of ‘‘spinning reserves’’ provided by gas turbines which oper- ergy storage to complement wind power in island locations with
ate very infrequently, at an enormous expense. Liu et al. [10] have isolated grids [13]. Of the large-grid studies, Swider [14] has con-
presented an analysis of increasing wind power in networks simi- cluded that the ‘CAES can be economic in the case of large-scale
lar to that of Victoria, and have concluded that the required in- wind power deployment’ in the German electricity grid. Ummels
crease in thermal generating capacity would severely restrict et al. [15] have concluded that ‘cost savings from energy storage in-
such a project. crease with the amount of wind power installed’ in the Dutch grid.
Electricity is bought and sold like any other commodity. As However, Lund et al. [16] have reported that CAES is not economic
such, its price depends on supply and demand. Power producers for the grid of Denmark. The above studies have shown that the lit-
submit bids at 5-min intervals – prices at which they are willing erature has indicated that economic feasibility of energy storage is
to sell proportions of their available power. The regulator, Austra- highly grid-dependent.
lian Energy Market Operator (AEMO), orders the bids, and sets the In contrast to the above studies, this paper investigates the eco-
price based on the highest bid which is required to meet the de- nomic feasibility of various energy storage systems to enhance
mand of the grid at a given time (meeting demand in the most cost wind power in a low penetration, base-load dominated, large
efficient way). The power producers which bid below that level capacity electricity grid. This project is an extension of the work
must provide power to the grid. The average price over six consec- by Hessami and Yeoh [17] and Hessami and Rocha [18], and it
utive 5-min intervals becomes the half-hour spot price; this is the incorporates more sophisticated optimisation, more detailed cost
price paid to power producers for the power which they provide analysis, an investigation of TES, and a method which allows for
over that half-hour period. reliable comparison between the different technologies.
Victoria has an electricity grid which is dominated by base-load The existence of PSHS and CAES systems in commercial opera-
thermal (coal fired) power stations. These stations run at very high tion indicates that they are certainly commercially viable under
capacity factors, and cannot be throttled easily. To meet the high the right conditions. However, it was not possible to find in the
demand caused by hot weather or other events, gas-fired peaking open literature an example of a commercially operating electri-
plants are used. These peaking plants run at capacity factors on cal–electrical storage system using TES as most studies utilise solar
the order of 2% [11], and as such, demand very high prices for heating or waste heat from industrial processes. However, a confi-
the power they produce. This makes the spot price in Victoria dential industry source has stated that a recent feasibility study
highly variable; prices can vary by several orders of magnitude about the use of TES has found it not to be feasible for their envi-
for small changes in demand. ronment and operating conditions. The model presented in this pa-
These market conditions create a significant opportunity. If the per will therefore serve to reinforce or counter their analysis.
output characteristics of a wind farm could be modified such that This research study incorporates technical feasibility, opera-
the energy is provided to the grid at times of high demand, rather tional optimisation, costing, and financial analysis elements as
than relying on wind, less peaking power stations would be re- shown in the block diagram of Fig. 2. A spreadsheet model which
quired, and the wind farm would become more profitable. was developed in this project as described in this paper is able to
Energy storage provides the ability to store power generated by optimise the parameters of the storage systems, as well as its oper-
wind turbines at times of low demand, and release that energy at ation, to determine the commercial viability of each storage system
times of high demand. This is clearly advantageous to the power in conjunction with the PWF application.
grid, and it also provides an economic incentive: wind power gen-
erators receive higher revenues by providing power to the electric-
2. Portland Wind Farm (PWF)
ity market when the spot price is high, as illustrated in Fig. 1. A
wind farm with energy storage can be classified as a ‘‘partially dis-
The PWF is consisted of four individual wind farms shown in
patchable’’ source because of its ability to respond to most situa-
Fig. 3; further details for each farm are provided in Table 1 [19].
tions of high demand.
PWF with a total installed capacity1 of 186 MW is located along
Large-scale energy storage is currently not in widespread use
the Victorian coastline, southwest of Melbourne. Sites along the
because the centralised nature of traditional electricity production
southern coast of Australia are ideal for wind power generation as
does not require it. Today, storage capacity worldwide is the equiv-
they experience consistently strong winds, are in close proximity
alent of about 90 GW of a total production of 3400 GW, or roughly
to the National Electricity Grid, have low population density, and
2.6% [3]. However, many storage options have been commercially
proven or investigated in the academic literature. Therefore, en-
1
ergy storage can ameliorate a wind farm’s inherent variability, The installed capacity of PWF in 2009 when this study was undertaken was
specified to be 194.4 MW. However, at that time, not all the wind farms were yet
and it can provide the ability to store power generated by wind completed. The revised total installed capacity of PWF is now 186 MW. All the
turbines at times of low demand, and release that energy at times calculations performed in this study are based on the original installed capacity of
of high demand. Wind power operators receive higher revenues by 194.4 MW.
M.-A. Hessami, D.R. Bowly / Applied Energy 88 (2011) 2755–2763 2757
Fig. 1. The connection of an energy storage system to this wind farm would allow energy to be stored when the wind output is high and the spot price is low, and to be
discharged when the spot price (that is, the power grid demand) is high.
Available Resource
Plant design
technology availability
Revenue
Return on
Investment
Fig. 2. Block diagram of the key information collected and the work completed over the course of this research project to model the wind power generation and storage
system to evaluate the economic viability of large-scale energy storage.
Table 2
Energy storage system parameters and capital costs. Costings and plant specifications are obtained from confidential industry quotations. Leakage losses are negligible for CAES
and TES over the timescales of interest. All costs are in 2008 Australian dollars (A$).
For the purpose of the model, the four wind farms of PWF are It should be noted that interest rate is not relevant when calcu-
treated as a single entity providing power to the Victorian grid. lating ROR. ROR is a measure of the ‘‘equivalent interest rate’’
The operation of the energy storage system was modelled using which would have been generated on an equivalent investment
historical operational data to determine the extra revenue that in, for example, a bank at that interest rate.
would be generated if energy storage was available. Three years
worth of Victorian electricity spot prices, in half-hourly intervals, 5. Operational strategy
were obtained and analysed. Three years of 5-min power readings
from Yambuk Wind Farm (one of the four wind farms making up The objective of this investigation is to find the maximum pos-
the PWF) were converted to half-hourly averages and scaled up sible revenue which can be generated by the addition of any given
to the output of PWF; note that the installed capacity of Yambuk storage system (that is, storage type and size as well as power rat-
is 30 MW compared to 194 MW for PWF. The standard deviation ing) over the 3 years of operation considered in the model. This re-
of the data conversion process was calculated to be 2.3%. The total quires determining the optimal operation for a given system on a
electrical energy output based on the half-hourly averages was real-time basis. For simplicity, it is assumed that the plant operates
found to be 94.8 GW h/year in 2007 which compares very favour- at the same conditions for the whole of each half-hourly period. If
ably with the ‘‘in excess of 90 GW h/annum’’ reported by the wind the storage is not close to being full or empty, the following oper-
farm’s owner and operator [12]. Both the data and the scaling pro- ations are possible:
cess were shown to be reliable by independent verification with
wind speed data from Port Fairy and Portland (see Fig. 3 for the 1. Storage is ‘‘charged’’ using power from PWF, selling no power
location of the wind farms as well as that of the weather stations) (or reduced power) to the grid.
obtained from the Australian Bureau of Meteorology over the same 2. Storage is ‘‘discharged’’, selling power from storage and wind
period. As shown in Fig. 5 for the 4000–4600 half-hourly interval of farm to the grid.
2007, the correlation between the wind speed data of Port Fairy 3. Storage is ‘‘bypassed’’, selling only power from the wind farm to
and the power output of Yambuk station is very good. A better cor- the grid.
relation is expected if actual wind data for Yambuk Wind Farm was
available and could have been used in the preparation of Fig. 5. There are many strategies available to operate the system to
The mathematical formulation and calculations of the computer generate extra revenue. Some of these were tested using the mod-
model were implemented in the Matlab computer software envi- el, and are explained briefly below. The revenues generated are
ronment. The model calculates the rate of return (ROR) or return compared in Fig. 6.
on investment for any system configuration. The number of tur- A simple technique is to charge the system whenever the spot
bines, and the storage size, can be varied. For maximum flexibility, price is below a ‘‘charge’’ (C) price, and discharge when above a
2760 M.-A. Hessami, D.R. Bowly / Applied Energy 88 (2011) 2755–2763
140
120
110
100
90
80
No storage Vary sell, fixed Vary sell, buy Account for Predictive c/d Improved Combination Optimum
buy storage level price predictive solution
Optimisation method
Fig. 6. The revenue generated using each operation strategy, compared with the revenue generated by the wind farm without storage. With increasing sophistication, the
techniques approach the optimum solution.
‘‘discharge’’ (D) price. An optimum C and D can be found by logic or parameters may be appropriate for each type and size of
exhaustive search. This technique is effective during normal oper- storage device. In order to reliably compare between different stor-
ation, but does not do well during periods of high or low spot price. age systems, the best possible revenue that can be generated by a
This strategy can be improved by modifying the C and D values given storage device over the 3 year data period needs to be found.
to Cm and Dm respectively, as shown in the following equation. Any A technique has been developed to achieve this.
storage level, S, below 1 (i.e., not full) increases the spot price at
which the system will begin to discharge, and thus obtain better 5.1. Dynamic programming
prices as the system discharges. W is a weighting factor which
determines how much of an effect the storage level is allowed to As mentioned above, sophisticated techniques can indeed gen-
have on the operation. The optimum modification function erate extra revenue as reflected in the diagram of Fig. 7. However,
(1 S)15, weighting factor W, and base values of C and D were to reliably compare between different storage systems, the best
determined by an exhaustive iterative search process. possible revenue should be found. Consider the ‘‘path’’ created by
a given combination of charge and discharge cycles shown in
Dm ¼ D ½1 þ W f ðSÞ; where f ðSÞ ¼ ð1 SÞ15 Fig. 8. If the revenue from the paths could be compared, the best
possible path could be selected. With three possible actions at each
The operation of the plant can be further optimised based on time step n, the number of possible paths is 3n, which quickly be-
the fact that the distribution of the spot price varies throughout comes too large to handle. However, using a method of dynamic
the year. To take this into account, an optimum C and D can be programming adapted from [16], it is possible to reduce the num-
determined using future prices for a given period, on the basis that ber of outcomes.
wind farm outputs and spot prices can be reliably forecasted under The only items of interest at the end of each time period are the
most conditions several days in advance. D is based on a top per- revenues generated up to that point, and the corresponding storage
centile, and C on a bottom percentile of prices. levels. If two paths ending at a given time have the same storage
It is clear that increasingly sophisticated operating logic can be level, only the path corresponding to the highest storage level
brought to bear in order to improve the performance of each stor- needs to be recorded; the other path is sub-optimal and can be dis-
age medium. The revenues generated by various operating strate- carded. By discretising the number of possible storage levels, it is
gies for a given PSHS system are shown in Fig. 7. However, different possible to make the storage levels overlap. This method clearly
ROR for these plant sizings
10 X: 1800
Y: 188.4
8 Z: 9.622
0
800
600 6000
5000
400 4000
3000
200 2000
1000
Output rating (MW) 0 0
Storage size (MWh)
Fig. 7. Results of optimisation calculations for the PSHS system. The designated point has the highest rate of return. The X, Y and Z coordinates are storage size (MW h), output
(MW) and ROR (%).
M.-A. Hessami, D.R. Bowly / Applied Energy 88 (2011) 2755–2763 2761
introduces some rounding error. However, the majority of this The two diagrams shown in Fig. 10 show the operation of the
rounding error is predictable: CAES system under different conditions. During periods of normal
Rounding error / discharge speed of device – Step size between spot prices (top diagram), the system charges and discharges on a
storage level discretisations daily basis, with the change in level over a day a function of the
Therefore, a subroutine was written to predict the rounding er- wind farm output and the spot price. In anticipation of a high spot
ror introduced by a given choice of discretisation in order to choose price on the bottom diagram of Fig. 10, the system is fully charged,
the number of discretisations to minimise the error introduced. and a large proportion of the system is discharged over the follow-
Testings showed that this code predicted the relative error with ing 12 h period.
great accuracy so that the correct step size could be chosen.
6.2. Sensitivity analysis
6. Results and discussion
A sensitivity analysis was performed to quantify the effect of
Simulations were undertaken for a wide range of storage sizes varying certain parameters on the calculated rates of return of each
and output ratings for each storage device, using the calculations system. The CAES system’s rate of return was found to be particu-
illustrated in the block diagram of Fig. 9 to find ROR for each con- larly sensitive to capital cost, fuel cost, and the annual increase in
figuration. The extra revenue gained by the ability to store more of electricity price. Variations in maintenance and running costs, and
the wind farm’s output, or to provide more power to the grid dur- charge and discharge efficiencies, had little effect on the profitabil-
ing high-demand periods, must be balanced against the increased ity of the system.
Fig. 9. Revenue is calculated by modelling the operation of the storage device for a given set of storage parameters. The rate of return on Investment is then calculated by the
model using the process shown above.
2762 M.-A. Hessami, D.R. Bowly / Applied Energy 88 (2011) 2755–2763
Table 3
Summary of optimum specifications for each plant, the capital expenditure required, and the revenues generated by each. All costs are quoted in 2007 Australian dollars (A$
million).
[9] Jaramillo OA, Borja MA, Huacuz JM. Using hydropower to complement wind [17] Hessami M-A, Yeoh D. A feasibility study of the use of seawater pumped hydro
energy: a hybrid system to provide firm power. Renew Energy storage on wind farms in Portland, Victoria, Australia. Wind Expo LAWEA
2004;29(11):1887–909. 2008, Mexico, 5–7 November, 2008.
[10] Liu J, van Kooten G, Pitt L. Integrating wind power in electricity grids: an [18] Hessami M-A, Rocha K. A study of an integrated compressed air energy storage
economic analysis, Department of Economics, University of Victoria, Canada; and large scale wind farm in Australia. Wind Expo LAWEA 2008, Mexico, 5–7
2005. November, 2008.
[11] NEMMCO. NEMMCO market data; 2009. <http://www.aemo.com.au/>. [19] Pacific Hydro. Portland wind energy project; 2010. <http://www.pacifichydro.
[12] Pacific hydro. Portland wind project; 2006. <http://www.pacifichydro.com.au/ com.au/en-us/our-projects/australia-/portland-wind-energy-project.aspx>
Default.aspx/tabid=134>. [accessed September 2010].
[13] Kaldellis JK, Kapsali M, Kavadias KA. Energy balance analysis of wind-based [20] Fujihara T, Imano H, Oshima K. Development of pump turbine for seawater
pumped hydro storage systems in remote island electrical networks. Appl pumped-storage power plant. Hitachi Rev 1998;47(5):199–202.
Energy 2010;87(8):2427–37. [21] IEA. Specifications of Okinawa seawater pumped storage power plant; 2006.
[14] Swider D. Compressed air energy storage in an electricity system with <http://www.ieahydro.org/01-Okinawa-Seawater-PSPP-lg.htm> [accessed
significant wind power generation. IEEE Trans Energy Convers 30.03.09].
2007;22(1):95–102. [22] Ridge energy storage. The economic impact of CAES on wind in TX, OK, and
[15] Ummels BC, Pelgrum E, Kling WL. Integration of large-scale wind power and NM: final report; June 27, 2005.
use of energy storage in the Netherlands’ Electricity Supply. IET Renew Power [23] Denholm P, Kulcinski G. Life cycle energy requirements and greenhouse gas
Generat 2008;2(1):34–46. emissions from large scale energy storage systems. Energy Convers Manage
[16] Lund H, Salgi G, Elmegaard B, Andersen AN. Optimal operation strategies of 2004;45(13–14):2153–72.
compressed air energy storage on electricity spot markets. Appl Therm Eng
2009;29(5–6):799–806.