1

New methodology for the generation of hourly wind speed data applied to the optimization of stand-alone systems
R. Dufo-Lpez and J. L. Bernal-Agustn, Member, IEEE
systems for stand-alone off-grid applications, the knowledge of wind speed hourly data series is very important to simulate and optimize the system, in order to assure the electrical supply of the system. In this work, we have improved the methodology shown in [2,3] to generate hourly wind speed series. The data needed to generate the hourly wind speed series are the 12 monthly average wind speed values (which can be obtained from internet sites as NASA website [4]) and the form factor of the Weibull distribution, which can be estimated. The results show that this methodology can achieve very good results. II. WIND SPEED HOURLY DATA GENERATION A. Data. The data used to generate the hourly wind speed data for the whole year (8760 values) are: a) 12 values corresponding to the monthly average wind speed, wav_m, where m is the month, 1 m 12. They can be obtained from wind atlas or NASA website [4]. b) Form factor of the Weibull distribution, bdist. It must be estimated. For example, in Spain, in the Ebro valley, a value of about 1.3 is adequate. c) Correlation factor, fc, usually values of about 0.85 to 0.95 are adequate. d) Optional data: If the dependency of wind speeds upon the time of the day is known, the following data must be estimated: 12 values of wn_m, the average night speed of the month; 12 values of Am, the amplitude of the month; 12 values of Fm factor of the month, which provides an indication of the dependency of wind speeds upon the time of the day; 12 values of hmax_m, the hour of the maximum speed of the day. With these values, for each month, the average speed of the hour h of the day (0 h 23), for the month m is (example in Fig. 1):

Abstract-- This paper shows a new methodology for the generation of hourly wind speed data used in the optimization of stand-alone systems. Wind speed hourly data is logged since many years ago in meteorological stations located near big cities or airports. However, in off-grid rural areas, where stand-alone systems such as wind-batteries or hybrid-wind-batteries systems can be installed, wind speed hourly data is not usually known. Monthly average wind speed data can be obtained from meteorological atlas or from some web pages as NASA website. The methodology presented in this paper shows the generation of the hourly wind speed data using as data just the 12 monthly average wind speed data values and the form factor of the Weibull distribution (which can be estimated from the values of the meteorological stations of the area). Index Terms Batteries, Data processing, Optimization, Power system simulation, Wind power generation.

IND speed hourly data is very important for the simulation and optimization of stand-alone windbatteries systems or wind-hybrid systems (usually wind-pvbatteries, wind-pv-diesel-batteries). However, hourly data series are not available for any location. These data are only available for locations near cities or airports, where meteorological stations have been logging data since many years ago. Stand-alone off-grid Wind-Batteries systems are not very usual, as hybrid systems including Photovoltaic or Diesel are more reliable and, usually, lest costly than the systems that use a single source of energy [1]. If the electrical load to be supplied is critical, wind-batteries systems are not the most suitable systems due to the variability of wind speed, this is why in these cases another source of energy must be added, generally Photovoltaic panels or Diesel genset, or both. These systems are known as hybrid systems. Even for non-critical electrical load, the use of hybrid systems is encouraged by many manufacturers of small wind turbines to be used in offgrid applications. Anyway, taking into account wind-batteries or hybrid-wind
R. Dufo-Lpez is with the Department of Electrical Engineering, Zaragoza University, Zaragoza, 50018 Spain (e-mail: rdufo@unizar.es). J.L. Bernal-Agustn is with the Department of Electrical Engineering, Zaragoza University, Zaragoza, 50018 Spain (e-mail: jlbernal@unizar.es).

I. INTRODUCTION

wm, h = wn_m + max[0, ( Am Fm(h hmax _m ) 2 )]

(1)

In this case, the average wind speed of the month, wav_m, can be calculated as follows:
wav _ m = wn_m +

Av_m Av_m Fm 18

(2)

2
6

Average wind speed of the month(m/s)

5 4 3 2 1 0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23

Second: we calculate the correlated values for each hour h of each day d of the year, cd,h, as follows: if ( d = 0 & h = 0) : cd , h = Weibulld , h (5)
else : cd , h

= cd , h 1 fc + Weibulld , h (1 fc)

(6)

Third: For each hour of the year, a fraction of the average value of the month is subtracted from each value of cd,h, getting the series ed,h: ed , h = cd , h - f subtract wav _ m (7 )
Hour of the day

Negative values will be converted in zero values: if (ed , h < 0) : ed , h = 0 (8) Fourth: For each month, we calculate the correction factor fcorrection_m so that the final value of the wind speed series have the same monthly average values as the data, wav_m: fcorrection _ m = wav _ m 24d m / ed , h (9 ) Where dm is the number of days of the month m. Fifth: Finally we scale the values ed,h getting the hourly wind speed wd,h so that the final series has the same monthly average values as the desired (wav_m). wd , h = ed , h f correction _ m (10 ) Sixth: We compare the cumulative distribution function of the series wd,h, F(wd,h), with the cumulative distribution function F(w) of a Weibul distribution with bdist form factor and average value the average wind speed of the year, wav_yr. For each interval of wind w in steps of 0.1 m/s, between 0 and 30 m/s, we compare the two distributions, and we calculate the sum of the differences in absolute values:
diff cum =
w =0

Fig. 1. Average wind speed of a month. The dependency of wind speed upon the time of the day is known. wn_m = 4 m/s; Am = 1 m/s; Fm= 0.05; hmax_m = 16.

Usually the dependency of wind speeds upon the time of the day is unknown, so optional data of d) is not known. Then, input data are only the values of average wind speed of the month, wav_m,, considering Av_m = 0 and wn_m = wav_m. Then, wm,h = wav_m. B. Methodology. In what follows, the steps of the algorithm to get the wind speed hourly data series are indicated. 1). Main for loop A main for loop is used to evaluate different form factors for the Weibull distribution b, from 0.3 to 5, in 0.1 steps (48 values are evaluated), in order to find the best form factor b which best fits the form factor desired for the final wind speed hourly data series that will be selected, bdist. The form factor b is used to generate, for each hour of the year, a number following a Weibull distribution with b form factor, shown later in (3). This series generated are processed later, so the final form factor of the series will be different from b. At the end, the series selected will be the one with the form factor same as the desired, bdist. 2). Secondary for loop For each value of b, a secondary for loop is used to evaluate different values fsubtract to subtract from the series a value proportional to the average of the month. The value of fsubtract varies from 0 to 1, in 0.1 steps (10 values are evaluated for fsubtract). 3). Calculations First: we calculate for each hour of the year (8760 hours) a number Weibulld,h following a Weibull distribution with b form factor, where d is the day (1 d 365) and h is the hour of the day (0 h 23). The values Weibulld,h of a month follow a Weibull distribution of b form factor and average value wav_m. For the same hour h of the days of the month, the values Weibulld,h follow the Weibull distribution with b form factor and average wind speed wm,h.
Weibulld , h = [ h ln(1 ad .h )]
b 1/ b

month

| F (w

30

d ,h

) F ( w) |

(11)

Where F(w) is calculated as follows: b F ( w) = 1 - exp(-(w/ ) dist ) (12 ) Where is the scale factor of the Weibull distribution with bdist form factor and average value the average wind speed of the year: = wav _ yr / (1 + 1 / bdist ) (13) 3). Selection of the best hourly wind series A total number of 48 (main loop for) x 10 (secondary loop for) = 480 series of 8760 hourly wind speed values for the whole year have been calculated. The series wd,h with b and fsubtract which fits best the Weibul distribution desired, with form factor bdist (i.e., the series for which diffcum presents the lowest value), is the one chosen. The values of the year corresponding to those values are stored. III. EXAMPLE OF HOURLY WIND SPEED GENERATION AND APPLICATION IN THE OPTIMIZATION OF A WIND-BATTERIES SYSTEM. An example of application of the method is shown in this section. We will compare the hourly wind speed hourly series data measured in the meteorological station of the Zaragoza airport during the year 2006 with the data generated synthetically as shown in section II.B.

(3)

Where and ad,h is a random number between 0 and 1, and h is the scale factor of the Weibull distribution for each hour of the day of the month, with b form factor: h = wm, h / (1 + 1 / b) (4) Where (1+1/b) is the Gamma function.

Weibulld,h

A. Hourly Wind speed measured in the Zaragoza Airport in 2006. Fig. 2 shows the wind speed series measured in the Zaragorza airport during year 2006. The wind speed distribution measured has an average value of 4.17 m/s and a Weibull factor of 1.3 (Fig. 3).
25

1000 900 800 700 600 500 400 300 200 100

20

Measured wind speed (m/s)

15

0 0 1000 2000 3000 4000 5000 6000 7000 8000

Hour of the year

10

Fig. 4. Weibulld,h: series of a Weibull distribution with b = 0.3 and average wind speed wm,h.
200 180

0 0 1000 2000 3000 4000 5000 6000 7000 8000

160 140

Fig. 2. Wind speed measured in the Zaragoza Airport, year 2006.

0.2 0.18 0.16 Measured Weibull dist. b = 1.3, average 4.17 m/s

cd,h 120
100 80 60 40 20 0 0 1000 2000 3000 4000 5000 6000 7000 8000

Probability density

0.14 0.12 0.1 0.08 0.06 0.04 0.02 0 0 2 4 6 8

Hour of the year

Fig. 5. Correlated values, cd,h.

30 25
10 12 14 16 18 20 22 24 26 28 30

wind speed (m/s)

Fig. 3. Probability distribution of wind speed measured in the Zaragoza Airport, year 2006.

wd,h (m/s)

20 15 10 5 0 0 1000 2000 3000 4000 5000 6000 7000 8000

B. Hourly wind speed generated. The data selected to generate the hourly wind speed are the 12 values corresponding to the monthly average wind speed of the wind speed measured in the Zaragoza Airport in 2006 (wav_m are 3.73, 4.89, 4.64, 4.97, 4.39, 4.37, 4.14, 5,98, 3.51, 3.24, 3.1, 3.09). The form factor used is bdist = 1.3 and the correlation factor chosen is fc = 0.9. Following the methodology shown in section II.B, the best hourly wind speed series has been generated. The best series found uses values b = 0.3, fsubtract = 0.8 and fcorrection_m = 0.17, 0.17, 0.17, 0.18, 0.15, 0.19, 0.14, 0.15, 0.2, 0.2, 0.21, 0.14. Fig. 4 to 6 show respectively series Weibulld,h, cd,h and the hourly wind speed series wd,h. Fig. 7 shows the probability distribution of wd,h. It can be seen that the distribution of the wind speed generated is similar to the distribution of the wind speed measured shown in Fig. 3. C. Optimization of a Wind-Batteries system using the measured wind speed data. We have used HOGA software [5], developed by the authors, to optimize a Wind-Batteries stand-alone system. The system must supply a low consumption domestic load of a house near Zaragoza Airport (3630 Wh AC load in one day).

Hour of the year

Fig. 6. Hourly wind speed wd,h.

0.2 0.18 0.16 Generated wind speed (wd,h) Weibull dist. b = 1.3, average 4.17 m/s

Probability density

0.14 0.12 0.1 0.08 0.06 0.04 0.02 0 0 2 4 6 8

10 12 14 16 18 20 22 24 26 28 30

wind speed (m/s)

Fig. 7. Probability distribution of wind speed.

DC bus is 48 V. Different types of DC wind turbines (0.4, 1.7 and 6.5 kW) and batteries (80 and and 444 Ah, 12 V) are considered in the optimization [3]. The number of possible combinations of components is: 2 models of batteries x 10 (batteries in parallel range between 1 and 10) x 3 models of wind turbines x 2 (wind turbines in parallel can be 1 or 2) = 120 possible combinations of components. We will use the enumerative optimization method (evaluating all the possible combinations of components) to find the optimal system (the combination which can supply the load with lowest Net Present Cost, NPC, including all the costs throughout the life of the system, 25 years). Using the measured data shown in section III.A, the optimal system includes 1 wind turbine of 6500 W and 4 serial x 10 parallel batteries of 80 Ah (total 38.4 kWh), with a total NPC of 29500 . The wind turbine generates 11,530 kWh in the year. Fig. 8 shows the simulation of the system (wind turbine output power and State Of Charge, SOC, of the batteries). In Fig. 8 we can see that the SOC is all the time higher than the minimum SOC allowed for the batteries, which is 40% of the maximum energy which batteries can store (38.40.4=15.36 kWh).
7 6 5 4 3 2 1 0 0 1000 2000 3000 4000 5000 6000 7000 8000 40 35 30 25 20 15 10 5 0

results of the optimization of wind-batteries or hybrid windbatteries systems using the methodology shown in this paper are quite good, in the majority of the locations studied the optimal system obtained is the same as the one obtained using measured wind speed data.
7 6 5 4 3 2 1 0 0 1000 2000 3000 4000 5000 6000 7000 8000 40 35 30 25 20 15 10 5 0

Wind turbine output (kW)

Hour of the year

Fig. 9. Simulation of the optimal system using wind speed generated data.

V. REFERENCES
[1] [2] M. Muselli, G. Notton, A. Louche, Design of Hybrid-Photovoltaic Power Generator, with Optimization of Energy Management. Solar energy, vol. 65(3), pp. 143-157, 1999. R. Dufo-Lpez, "Dimensionado y control ptimos de sistemas hbridos aplicando algoritmos evolutivos," Ph.D. dissertation, Dept. Electrical. Eng., Univ. Zaragoza, 2007. < http://www.unizar.es/rdufo/Tesis_Rodolfo_Dufo.pdf> R Dufo-Lpez, J.L. Bernal-Agustn, J. Lujano and J.A. DomnguezNavarro, " Utilization of synthetically generated hourly wind speed data in the optimization of Wind-Batteries stand-alone systems," in Proc. 2011 International Conference on Renewable Energies and Power Quality <http://www.icrepq.com/icrepq%2711/297-dufo.pdf> Surface meteorology and Solar Energy A renewable energy resource web site (release 6.0) sponsored by NASA's Earth Science Enterprise Program: <http://eosweb.larc.nasa.gov/sse/RETScreen/> Dufo-Lpez, R., Bernal-Agustn, J.L.,. HOGA (Hybrid Optimization by Genetic Algorithms) software. Free software available at: <http://www.unizar.es/rdufo>

Wind turbine output (kW)

Batteries SOC (kWh)

[3]

[4]

Hour of the year

[5]

Fig. 8. Simulation of the optimal system using wind speed measured data.

D. Optimization of a Wind-Batteries system using the generated wind speed data. We have optimized the same system shown in section III.C using the synthetically generated hourly wind speed series of section III.B. The optimal system found by HOGA is the same as the one found using the measured data. The wind turbine generates 11,161 kWh in the year. Fig. 9 shows the simulation of the system. IV. CONCLUSIONS The methodology shown in this paper to generate synthetically hourly wind speed series gives good results for the optimization of wind-batteries and hybrid-wind-batteries stand-alone systems. Data required is the average wind speed of each month of the year (which can be obtained, for example, in [4]), the Weibull form factor (which can be estimated depending on the region) and a correlation factor (getting good results using 0.9 for many locations in the world). Comparing with the methodology shown in [2,3], the

VI. BIOGRAPHIES
Rodolfo Dufo-Lpez obtained the Ph.D. degree in Electrical Engineering from the University of Zaragoza (Spain) in 2007. He worked as a teacher in secondary schools from 1998 to 2005. From 2005 to 2008 he worked as engineer. Simultaneously since 2004 he was part time lecturer in the Electrical Engineering Dept. of the University of Zaragoza. Since 2010 he is Profesor Ayudante Doctor at the University of Zaragoza.

Jos L. Bernal-Agustin (M04) received his Ph.D. degree in Electrical Engineering from the University of Zaragoza (Spain) in 1998. In 1991, he joined the University of Zaragoza, where he is, currently, Profesor Titular. He has been involved in research projects referred to design of power distribution network and renewable energy sources.

Batteries SOC (kWh)