IEEE TRANSACTIONS ON ENERGY CONVERSION, VOL. 19, NO.

2, JUNE 2004

441

Energy Storage and Its Use With Intermittent Renewable Energy

John P. Barton and David G. Infield
AbstractA simple probabilistic method has been developed to predict the ability of energy storage to increase the penetration of intermittent embedded renewable generation (ERG) on weak electricity grids and to enhance the value of the electricity generated by time-shifting delivery to the network. This paper focuses on the connection of wind generators at locations where the level of ERG would be limited by the voltage rise. Short-term storage, covering less than 1 h, offers only a small increase in the amount of electricity that can be absorbed by the network. Storage over periods of up to one day delivers greater energy benefits, but is significantly more expensive. Different feasible electricity storage technologies are compared for their operational suitability over different time scales. The value of storage in relation to power rating and energy capacity has been investigated so as to facilitate appropriate sizing. Index TermsEnergy storage, interconnected power systems, modeling, power distribution, voltage control, wind power generation.

I. INTRODUCTION

NERGY storage systems installed within an electricity system can be provided by a range of technologies and can add value in a variety of ways as summarized in Table I. Income may be derived from an energy store by charging it when the local electricity value is low and discharging it when the value is high. If, at some times, the grid at the point of connection of the embedded renewable generation (ERG) cannot absorb the entire output of the generator, then output must be curtailed and the value of the excess is effectively zero. If, at other times, demand is high and expensive generators are used to meet that demand, then the price will be high. Another source of income results from supplying ancillary services, for example, reactive power, voltage, and frequency control and emergency power during a power outage. This paper focuses on the first possibility identified, in which the value of a store depends on time variations in the local cost of electricity. In this study, the local grid voltage has been modeled in detail, but the impact on transmission losses has not. The costs of a store comprise capital costs, maintenance costs, any cost penalty associated with possible technical failure of the store, and the cost of electricity losses (both standing losses and energy transfer losses). In this paper, electrical losses have been modeled by a simple, overall round-trip efficiency, calculated at a typical time scale for the particular storage technology.

Wind power is the fastest-growing renewable energy source and poses the most immediate grid connection problems; the time variable nature of wind power is well documented. For these reasons, this paper focuses specifically on the use of energy storage with wind power and from this point on, ERG can be taken to be wind-powered. The output from a wind turbine depends on the wind speed, which varies across a wide range of time scales. The Van Der Hoven spectrum [1] describes variations up to 1000 h. Corresponding spectra must be calculated from local wind speed data in order to predict accurately the character of variations in wind power at a given location, required for the design of a suitable energy store. The loads on an electricity supply system, which are reflected in the spot price of electricity, vary according to the time of day, day of the week, season, and also other, unpredictable factors. The latter, essentially stochastic, elements are minor in comparison to wind power fluctuation, and have been ignored in this study. The amount of ERG that an electricity grid can absorb at a particular location is usually limited by voltage rise. The voltage drop caused by local loads offsets the voltage rise of ERG and, therefore, higher levels of ERG are permitted when local loads are high. In general, this tends to happen when total system electricity demand is high and when wholesale electricity prices are high. Short-term energy storage will smooth out the effects of wind turbulence on the power output from ERG and so facilitate a higher installed capacity, without exceeding local voltage limits. Longer-term energy storage (i.e., over several hours) could time-shift ERG power to a time when the local loads are higher and, therefore, allow even more generation capacity. The absorption of reactive power reduces the local voltage but the injection of reactive power increases the local voltage [2]. Energy storage may be an economically favorable alternative to allowing increased local reactive power loads (or rather, not reducing these); a strategy sometimes used to limit the voltage rise caused by ERG. However, a study of reactive power control is outside the scope of this paper. II. MODELING ASSUMPTIONS A. Storage Technologies Table II summarizes the key properties of each storage technology. The capital costs have been discounted at an annual interest rate of 5% to give annual costs. This reflects typical current interest rates. Note that in the case of hydrogen and redox flow cells, the energy rating of the store in kilowatt-hours can be chosen independently of the power rating of the store in kilowatts, as indicated by a Y in the third column. For these technologies, when designing a store for a given time scale, the

Manuscript received October 25, 2002. The authors are with the Centre for Renewable Energy Systems Technology at the Department of Electronic and Electrical Engineering, Loughborough University, Leicestershire LE11 3TU, U.K. (e-mail: j.p.barton@lboro.ac.uk). Digital Object Identifier 10.1109/TEC.2003.822305

0885-8969/04$20.00 2004 IEEE

442

IEEE TRANSACTIONS ON ENERGY CONVERSION, VOL. 19, NO. 2, JUNE 2004

TABLE I STORAGE TECHNOLOGIES AND APPLICATIONS

TABLE II PROPERTIES OF STORAGE TECHNOLOGIES

Fig. 1. Costs of energy storage technologies.

cost of the energy storage equipment is added to the cost of the power conversion technology. In all other technologies, the energy rating of the store is closely linked to its power rating, as indicated by an N in the third column. In this study, the size and consequent cost of these stores is estimated on the basis of the larger of the two components: the required power rating or the required energy rating at the given operating time scale. We define a characteristic time scale for each technology as the typical time taken to discharge the store. The charging time may be longer in some cases. The data presented in Table II have been derived from journal articles and websites, many from a particular edition of IEE

Power Engineering Journal [3]. Most are the projected costs expected within five years since the technologies are not yet widely available. The exceptions are for superconducting magnetic energy storage (SMES), lead-acid batteries, and nickelcadmium batteries that are all based on available costs for current technology. Total annual costs and energy efficiencies have been used to calculate a cost of storage per kilowatt-hour discharged from the store (Fig. 1). The costs of the electrical losses have been added at a typical U.K. wholesale price of 0.05 /kWh. Only the lowest cost technologies have been included in Fig. 1. Only one method of storing hydrogen has been evaluated: hydrogen is produced by an electrolyzer, stored in a high-pressure tank and used in a proton exchange membrane (PEM) fuel cell. For clarity, lead-acid batteries have been omitted from Fig. 1. Their cost per kilowatt-hour is very similar to that of sodium sulphur batteries at time scales less than 1 h while at longer

BARTON AND INFIELD: ENERGY STORAGE AND ITS USE WITH INTERMITTENT RENEWABLE ENERGY

443

TABLE III MODEL OF NATIONAL ELECTRICITY DEMAND AND LOCAL NETWORK CAPACITY

Fig. 2. Variations in system electricity demand.

Fig. 3. Load-price curve of the England and Wales Electricity System.

time scales, they are more expensive than sodium sulphur and nickel-metal-hydride batteries. B. Network Effects System electricity demand data have been taken for the electricity system of England and Wales [4]. This paper assumes that the electricity demand is a simple function of time of day and season. The year has been divided into two equal periods: summer and winter. The day is then divided into three power levels of peak, mid, and offpeak demand. Fig. 2 shows the daily load profiles and the three-level models used in calculations. The demand data [4] have been combined with price-per-unit data from the electricity pool in 1999 [5] to produce a system price-demand curve, Fig. 3. The grid connection to be modeled is based on a typical location in the 11-kV electricity grid of a rural area of the U.K. The chosen point for injection of wind power has a relatively low fault level of 12 MVA and a low X/R ratio of 0.46. The fault level is very similar to that used in a recently published analysis of a weak grid [6]. The grid connection presented in this paper has a local node voltage under maximum winter loads of 0.92 p.u. The electricity supply regulations permit voltage variations not exceeding % from the nominal 11 kV [7]. However, this limit must be maintained for all locations in the network. It is likely that the permitted local voltage change caused by ERG is smaller than this. The model presented in this paper assumes a maximum permitted voltage of 0.98 p.u., representing a 6% rise under maximum winter loads. For simplicity, the local loads are assumed to be proportional to the average national load although this is only approximately true. The local network is believed to have a greater proportion of domestic and agricultural premises than the national average. Their power consumption changes
Fig. 4. Composite 1-MW turbine power curve from three commercial turbines.

greatly with the time of day. In contrast, the national load contains more industrial users, whose consumption is more nearly constant. The voltage drop before the addition of any ERG is assumed to be proportional to the national load at the time in question. The approximate voltage rise due to ERG at the node then depends on the injected power, (1), [2] (1) where voltage rise caused by the ERG; nominal line voltage, 11 kV; injected active power; line resistance; injected reactive power; line inductive reactance. were calculated from knowledge of The values of and the local network. The load profiles of Fig. 2, price curve of Fig. 3, and the above formula have been used to model the behavior of the electricity pool and the example grid connection, as shown in Table III. C. Wind Turbine Modeling A generic power curve for a 1-MW wind turbine has been used for the calculations presented here, Fig. 4. The curve was constructed from data for commercially available wind turbines

444

IEEE TRANSACTIONS ON ENERGY CONVERSION, VOL. 19, NO. 2, JUNE 2004

TABLE IV ILLUSTRATION OF AN HOURLY SPREADSHEET METHOD

Fig. 5. Wind speed variation spectra adjusted to 8-m/s mean wind speed and 24% turbulence intensity.

[8]. This generic turbine has a cut-in speed of 3 m/s and a rated speed of 13 m/s. It has a hub height of 60 m, at which the annual average wind speed for the site in question is taken to be 8 m/s. The power output was scaled to model different levels of wind turbine capacity. D. Wind Speed Variations Hourly wind speed data from Cottesmore, U.K., has been scaled to 8 m/s and used to construct a probability distribution of wind speeds. To a good approximation, this is a Weibull distribution with a shape factor of 2.0 (a Rayleigh distribution). This still reveals nothing directly about the time scale of variations. The Van Der Hoven spectrum was therefore used and scaled to contribute the same total variation in hourly wind speed as the Cottesmore data, Fig. 5. The wind speed data from Cottesmore have also been used to construct an approximate spectrum of wind speed variations on all time scales down to 1 h. This is compared with the scaled Van Der Hoven spectrum in Fig. 5. Since the Cottesmore wind speed data only measure wind speed variations down to a time scale of 1 h and the Van Der Hoven spectrum is only indicative for the turbulence component, the turbulent variations have been calculated from a theoretical spectrum similar to the Von Karman spectrum [1], with a turbulence intensity of 24%. We can see that the very low frequencies identifiable from the Cottesmore data are of limited significance. The turbulence spectrum has been grafted onto the end of the Van Der Hoven spectrum, Fig. 5, and this combined spectrum is therefore used in what follows. III. CALCULATION APPROACH A. Probabilistic Method The operation of energy storage has been modeled using probabilistic methods; these have been incorporated into two spreadsheet calculations. The first spreadsheet models the operation of storage hour-by-hour, while the second models over one-day periods. The spectral gap in the middle of the Van Der Hoven Spectrum, Fig. 5, means that the distribution of hourly average wind speeds is almost identical to that of 10-min average wind speeds. The first spreadsheet is therefore also fairly representative of energy storage over 10 min. A very simplified example of an hourly spreadsheet is shown in Table IV, but the daily spreadsheet is organized in a similar

way. In the example spreadsheet, the long-term average wind speed is 8 m/s and the standard deviation of wind speed variation within each hour is 1.95 (representing a turbulence intensity of 24%). The distribution of hourly average wind speeds is modeled by a Rayleigh distribution. Each row represents a different interval of average wind speed over a period (e.g., 1 h). Let us call the midvalue of this interval . In the example, takes three possible values: 4 m/s (representing hourly-average wind speeds from 0 to 6 m/s); 8 m/s (representing hourly-average wind speeds from 6 to 10 m/s); and 12 m/s (representing hourly-average wind speeds above 10 m/s). Each element within each row represents a different interval of instantaneous wind speed (second by second) higher or lower than . The instantaneous , where is the interval midvalue of the instantaneous variation. The matrix has been constructed so that each column represents a fixed ratio of (for the case of hourly modeling, this represents a fixed value of turbulence intensity). In the example spreadsheet of Table IV, takes values of and times the standard deviation of variation within 1 h. For example, when the hourly average takes values of about m/s, 0 m/s, and m/s, is 4 m/s, giving absolute wind speeds of 3 m/s, 4 m/s, and 5 m/s. These actually represent ranges of wind speeds: values of the first column represents values of wind speed less than the m/s); mean minus half a standard deviation (

BARTON AND INFIELD: ENERGY STORAGE AND ITS USE WITH INTERMITTENT RENEWABLE ENERGY

445

the middle column represents wind speeds from mean minus half a standard deviation to mean plus a half standard devia; the final column represents tion wind speeds greater than mean plus half a standard deviation . Each element in the matrix is also associated with a wind turbine power output calculated using the turbine power curve of , see Table IV. For each Section II-C at a wind speed of time scale considered, all variations in wind speed slower than the time scale in question are described by one probability distri, (one value of probability for each row of bution function, the matrix). Variations faster than the time scale are described by a second probability distribution of wind speeds within the time , (one value for each column of the matrix). In the period, case of a 1-h store, this second distribution describes turbulent variations in wind speed, and is best represented by a Gaussian distribution. In the case of a daily store, this second distribution would include turbulent variation and weather system (synoptic) variation up to periods of one day. The wind speed distributions are described in more detail in Section III-B below. The wind speed spectrum is thus split into two, at 1 h or 1 day, Fig. 5. To the right of the appropriate line in Fig. 5, all variation in power output is assumed to be smoothed by the energy store. To the left of the line, the variations cannot be smoothed by the store. In the spreadsheets, energy can be taken from points with high wind power output and given to other points in the same row with lower power. This represents the action of a store absorbing transient surpluses and delivering this as useful energy a short period later. Energy cannot be exchanged between rows because this would require longer-term energy storage than what is being modeled. The precise redistribution is not important and is not calculated, but the spreadsheets do calculate the total amount of power transferred and the total amount of extra energy exported to the electricity network, see Table IV. For the example in question, the chosen maximum charging and discharging power of the store is 300 kW. One particular time of day has been chosen, in which the maximum power that can be absorbed by the local electricity network is 602 kW (The spreadsheet must be run again for other times of day when the local electricity demand is different). Let us now work through each hourly-average case. 1) Row 1: When the hourly average wind speed is 4 m/s, the wind turbine is only generating for a small proportion of the time. When it does generate, the local network can absorb the entire power. The store would be able to export an additional 300 kW (limited by the power rating of the store) but this power is just not available. In this case, the store does not increase the power exported. is 8 m/s, 2) Row 2: When the hourly average wind speed there are occasions when the wind power exceeds 602 kW (by 39 is 10 m/s). The probability-weighted average kW when excess power is 12 kW, but the store has a finite round-trip efficiency of 90%, so the power available is only 11 kW. The store would be able to export an additional 185 kW on average (sometimes limited by the network capacity and sometimes by the store discharge power rating of 300 kW). However, the extra exported energy is limited by the former quantity, the available stored power. Therefore, the extra exported energy is only 11 kW.

is 3) Row 3: When the hourly average wind speed 12 m/s, the wind power exceeds the local network capacity most of the time. The wind power exceeds the limit of 602 kW by 243 kW on average. The average excess power that the store could absorb is then 207 kW, limited by the store maximum charging rate of 300 kW. When multiplied by the store efficiency of 90%, this becomes 186 kW. However, on this occasion, the power that the store can export is limited by the local network capacity. The average extra exported power is therefore only 28 kW. Of course, the actual spreadsheets include many more values (rows) and (columns) to achieve a representative of model. The spreadsheets include the round trip efficiency of an appropriate energy storage technology. The spreadsheet for 1-h storage assumes an efficiency of 90% appropriate to a high-speed flywheel system. The Daily spreadsheet assumes an efficiency of 70% appropriate to an electrochemical store, for example, the Regenesys flow cell system. B. Wind Speed Distributions All wind speed distributions were corrected to a long-term mean wind speed of 8 m/s. At short time scales (e.g., 1 h), the wind speed is described by a long-term Weibull distribution of hourly average wind speeds combined with short-term, Gaussian distributions of turbulent wind speeds, about the hourly average. The Cottesmore data were used to directly calculate the Weibull shape factor of 2.0 for the distribution of hourly wind speeds (Section II.D). At the time scale of one day, the wind speed variation may be described by a combination of the long-term Weibull distribution of daily averages and a short-term Weibull distribution of instantaneous wind speeds within each day. In this study, the distribution of daily average wind speeds has been calculated from the area under the Van Der Hoven Spectrum from very long time scales up to daily time scales. A Weibull distribution with a shape factor of 2.92 gives the correct variance, and is also a good approximation to the measured distribution of daily average wind speeds at Cottesmore, when corrected to an average wind speed of 8 m/s. The distribution of wind speeds within each day has then been approximated by a Weibull distribution with a shape factor of 2.40 in order to give the correct total variance of wind speeds. This total variance was calculated by adding the variance in the Cottesmore hourly wind speed data to the variance of short-term turbulence. C. Local and National Network Effects Wind power offsets some local electricity demand. The level of power that can be absorbed by the local network depends linearly on this local electricity demand, which is assumed to be proportional to the national total demand (Section II-B). The price also changes with the national demand. All of these variations are treated as deterministic. The 1-h spreadsheet has been run separately for each level of network load to represent the different times of day and season. The daily storage spreadsheet, however, sums the total stored and exported energy for all three levels of network load within one day (peak, mid, and offpeak), weighted for the number

446

IEEE TRANSACTIONS ON ENERGY CONVERSION, VOL. 19, NO. 2, JUNE 2004

of hours in the day for which each level applies. Thus, in the daily spreadsheet, energy is transferred between different levels of network load as well as between different levels of turbine power output. The daily spreadsheet also calculates the revenue gained, based on the prices in each time of day and season. The method of calculation is otherwise very similar to that described in Section III-B, and is run once for a typical summers day and once for a winters day. D. Control Strategies In addition to the power rating and energy capacity, the mode of operation (control strategy) must be decided. Two potential strategies have been compared with each other and with the option of simple power curtailment, all described below. The store operator has a choice whether to release energy from a store as soon as the local network can absorb it, or whether to store more energy and wait until the price of electricity is higher before releasing it. The strategies have been evaluated using the probabilistic spreadsheets of Section III-A. The difference in the price of electricity has been balanced against the finite energy efficiency of the store. 1) Power Curtailment: Curtailment involves temporarily reducing the output power of wind turbines or directing some of their power to a dump load. The power exported is limited to the maximum power that does not cause the local network voltage to exceed its limit. This varies with the time of day and season, Table III. Even if an energy store with a high power rating is installed, if the wind power capacity is higher than 261 kW, there are occasions when curtailment is still necessary because accumulated excess power output exceeds the energy capacity of the store. Energy curtailment is therefore also used in combination with the other options of Sections II and III below. 2) Maximizing Energy Export: In this strategy, illustrated in Section III-A and Table IV, energy is stored only when the voltage would otherwise exceed the maximum allowed. Again, the maximum power exported is set by the voltage limit. Energy is exported as soon as the wind power drops below this maximum power. This strategy minimizes the total amount of energy entering and leaving the store and so reduces losses due to the finite round-trip efficiency. 3) Power Leveling (or Net Load Leveling): The sum of local generation and local loads is kept as constant as possible. This means that the local grid voltage is held very nearly constant and transmission losses are minimized. The exported power is set to either the average net power during the period, or the maximum determined by the voltage limit, whichever is the smaller. The net power is determined by the difference between the wind power and the maximum that the local network can absorb. Using the example of Table IV again, when the hourly average wind speed is 8 m/s, the average wind power is 380 kW but the network can absorb a maximum power of 602 kW. The network could absorb an extra 222 kW during this hour. The store is operated to supply a constant power to the grid of almost 380 kW. More power is being stored and discharged than in the strategy of maximizing energy export, above. The round trip efficiency of 90% therefore has an increased effect. The average power exported to the grid is only 372 kW. Now, 8 kW is lost in the store

TABLE V POWER RATING OF A STORE

compared to only 1 kW when employing a strategy of maximum energy export. The energy lost due to the round trip efficiency limits of any practical store is much greater than the transmission losses saved. The grid is a more efficient battery than the store, so it is not worth operating a 10-min or 1-h store in power leveling mode. Power leveling may be worth doing with daily storage because changes in electricity prices can be exploited. The calculation method for daily power leveling is almost the same as illustrated above for hourly storage, but the power absorbed by the network changes with the time of day, so the calculation of net power is longer. E. Power Rating Calculations The spreadsheets developed on the basis of the probabilistic approach were used to optimize the power rating of a wind farm and the power rating of an energy store connected to it. The probabilistic spreadsheet method predicts that compared with a store that delivers the maximum possible extra energy to the grid, a store with half of the power rating still delivers 90% of this benefit, see Table V. The power rating of the store increases only slightly between 1-h and 24-h storage. F. Energy Rating Calculations The probabilistic spreadsheet method cannot capture the temporal aspects of operation: when and how often the store fills and empties, and hence it cannot predict the required energy rating of the store. The impact of energy capacity on system performance is the subject of ongoing research. However, a first estimate of energy capacity can be made for a given power rating using the cost curves of Fig. 1 and assuming that the store fills and empties in a sinusoidal manner Energy capacity Power Time Hours

Some results are presented in Section IV-D below. IV. RESULTS The spreadsheet calculation method predicts energy outputs as shown in Fig. 6 and revenue earned as shown in Fig. 7. The different control strategies are compared. The power output from the turbine and its potential economic value give a straight line on each graph, representing the calculated load factor of 0.39. A. Power Curtailment The local grid as modeled here can always absorb all wind power up to a capacity of 261 kW. As the wind power capacity

BARTON AND INFIELD: ENERGY STORAGE AND ITS USE WITH INTERMITTENT RENEWABLE ENERGY

447

rating of 300 kW. If the store fills and then empties in a sine wave, the energy capacity is

Fig. 6.

Annual energy exported to the grid.

This would cost per year and would gain per year in extra revenue. Fig. 1 shows that the cheapest hourly energy storage is also a flywheel system. The longer time requires a larger store than 10-min storage. However, little additional revenue is gained compared to 10-min energy storage because the distribution of hourly mean wind speeds is very similar to that of 10-min average wind speeds. Daily energy storage is most economically achieved using a Regenesys plant. The power rating is 400 kW and the required kWh. This would cost energy capacity is per year. This would gain per year in extra revenue. only V. CONCLUSION This research has established that key aspects of the use of energy storage can be captured by a simplified probabilistic approach, which requires only limited input data. Specific conclusions based on the modeling indicate the following. 1) Energy storage over 10 min using, for example, flywheels allows 10% more wind energy to be absorbed without grid reinforcement and appears to be economically worthwhile. 2) Energy storage over 24 h using redox flow cells allows up to 25% more wind energy to be absorbed and 30% more revenue to be earned but does not appear to be economically justified. 3) Energy storage up to one day cannot exclude energy curtailment in a weak grid connection. 4) Energy management incorporating energy storage over 24 h and energy curtailment can allow up to three times the amount of wind energy to be absorbed by a weak grid compared to conventional grid connection of wind farms. The impact of energy capacity has been sidestepped in this study through the assumption of a storage capacity that is large enough for the time scale considered. Nevertheless, the conclusions should be valid provided reasonably generous storage capacities are used of the orders of magnitude estimated in Section IV-D. The issue of capacity is complex and, as already mentioned, is the subject of ongoing research. ACKNOWLEDGMENT The authors of this paper would like to express their gratitude and sincere appreciation to R. Cooke of Alstom, and Dr. S. Watson and M. Thompson of CREST for their valuable comments and suggestions. We would also like to thank the British Atmospheric Data Centre for supplying wind speed data. REFERENCES
[1] U. Hassan and D. M. Sykes, Wind structure and statistics, in Wind Energy Conversion Systems, L. L. Freris, Ed. Englewood Cliffs, NJ: Prentice-Hall, 1990, ch. 2, pp. 1132.

Fig. 7.

Annual revenue earned from electricity.

is increased, the fraction curtailed remains small until about 500 kW when the exported energy flattens out. The fraction of revenue lost is not as great as the energy lost since most of the curtailment occurs when the price of electricity is low. B. Maximizing Energy Export This strategy, with either 10-min or 1-h storage, results in an extra 10% of energy and revenue compared to power curtailment alone. Maximizing energy export with daily storage gains an extra 30% energy but only 25% revenue. This is because the extra energy is stored for the minimum length of time, and exported mainly at times when the price is still low. C. Power Leveling (or Net Load Leveling) Even without a wind turbine, the store gains revenue by importing electricity from the network at night and exporting electricity during the day. With a 1-MW capacity wind turbine, the extra energy exported is only 25% due to the higher energy losses of the store. However, the extra revenue is 30% because more electricity is exported at times of high price. D. Cost-Benefit Analysis This cost-benefit analysis is based on the results of the probabilistic spreadsheets, Fig. 1 and Table II. These costs of storage are in addition to the cost of the wind turbines. Fig. 1 shows that 10-min energy storage is most economically achieved using a flywheel system. The store requires a power

448

IEEE TRANSACTIONS ON ENERGY CONVERSION, VOL. 19, NO. 2, JUNE 2004

[2] N. Jenkins, R. Allan, P. Crossley, D. Kirschen, and G. Strbac, Technical impacts of embedded generation on the distribution system, in Embedded Generation, London, U.K.: The Institution of Electrical Engineers, 2000, pp. 1112. [3] J. N. Baker, A. Collinson, G. W. Hunt, A. Green, A. Price, S. Bartley, S. Male, G. Cooley, A. Shibata, K. Sato, E. Kodama, Y. Kurashima, P. Lex, B. Jonshagen, R. Schottler, R. G. Coney, A. Campbell, R. McHattie, C. Tarrant, and O. Weinmann, Special feature: Electrical energy storage, Inst. Elect. Eng. Power Eng. J., vol. 13, no. 3, pp. 107180, June 1999. [4] I. P. Burdon, Options for mid-merit power generation in the UK electricity market, Inst. Elect. Eng. Power Eng. J., vol. 12, no. 3, pp. 115122, June 1998. [5] D. Milborrow, Revolutionary potential, Wind Power Monthly, vol. 16, no. 10, Oct. 2000. [6] J. Mutale, G. Strbac, S. Curcic, and N. Jenkins, Allocation of losses in distribution systems with embedded generation, in Proc. Inst. Elect. Eng., Gen. Transm. Dist., vol. 147, Jan. 2000, pp. 714. [7] Quality of Supply [Online]. Available: http://www.nationalgrid.com/ uk/activities/mn_quality.html [8] Bundesverband Windenergie e.V., Windenergie 2000, A wind turbine catalogue, ISBN 3-9 806 657-2-0, 2000.

John P. Barton was born in Nottingham, U.K., on September 20, 1966. He received the B.A. and M.A. degrees in mechanical engineering from Cambridge University, Cambridge, U.K., in 1989 and 1993, respectively. He received the M.Sc. degree in renewable energy systems technology from Loughborough University, Leicestershire, U.K., in 2001, where he is currently pursuing the Ph.D. degree. He worked on the design of gas turbine compressors and fans at Rolls-Royce plc. from 1990 to 2000. His area of interest is the use of energy storage with renewable energy sources in electricity networks.

David G. Infield was born in Paris, France, on March 30, 1954. He received the B.A. in mathematics and physics from the University of Lancaster, Lancaster, U.K., and the Ph.D. degree in applied mathematics from the University of Kent at Canterbury, Canterbury, U.K. Currently, he is Director of the Centre for Renewable Energy Systems Technology (CREST) and Professor of Renewable Energy Systems with the Department of Electronic and Electrical Engineering at Loughborough University, Leicestershire, U.K. He was with Building Services Research and Information Association, Bracknell, U.K., working on solar thermal system design. He was also with the Rutherford Appleton Laboratory, Didcot, Oxfordshire, U.K., from 1982 to 1983, working on wind energy systems and electricity supply modeling.