Optimal Placement of Wind Turbines in A Wind Park Using

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 6

ARTICLE IN PRESS

Renewable Energy 33 (2008) 14551460 www.elsevier.com/locate/renene

Optimal placement of wind turbines in a wind park using Monte Carlo simulation
Grigorios Marmidis, Stavros Lazarou, Eleftheria Pyrgioti
High Voltage Laboratory, Department of Electrical and Computer Engineering, University of Patras, 26500 Rio, Patras, Greece Received 15 May 2007; accepted 7 September 2007 Available online 22 October 2007

Abstract In the present study, a novel procedure is introduced for the optimal placement and arrangement of wind turbines in a wind park. In this approach a statistical and mathematical method is used, which is called Monte Carlo simulation method. The optimization is made by the mean of maximum energy production and minimum cost installation criteria. As a test case, a square site is subdivided into 100 square cells that can be possible turbine locations and as a result, the program presents us the optimal arrangement of the wind turbines in the wind park, based on the Monte Carlo simulation method. The results of this study are compared to the results of previous studies that handle the same issue. r 2007 Elsevier Ltd. All rights reserved.
Keywords: Wind parks; Wind energy; Monte Carlo simulation; Optimization

1. Introduction Nowadays the production of electrical energy from renewable sources is not just a case of study but an irreplaceable ring in the chain of production. Already a tidy sum of the total global production is taking place on wind farms all over the world. Of course, we are just at the beginning, as an increase of production of energy from renewable energy sources is expected. Many countries have already invested on green power and they will still invest more due to the diminishing fossil fuel resources, the Kyoto commitment and the obligations that result from it for every country with regard to the protection of the environment. Goals like the recent one that EU has decreed that 20% of European energy production must be produced from RES till 2020 looks tough today but not impossible, if all countries stick to their commitments. In the present scenes wind energy has a crucial role in this RES evolvement, as it has the vast part of the total RES energy production. Nevertheless, wind energy production
Corresponding author. Tel.: +30 2610 996816; fax: +30 2610 997358.

E-mail address: grmarmidis@hvlab.ee.upatras.gr (G. Marmidis). 0960-1481/$ - see front matter r 2007 Elsevier Ltd. All rights reserved. doi:10.1016/j.renene.2007.09.004

is so rapidly expanding that only in EU we had a 23% increase last year, in comparison to the year 2005, of wind power capacity according to the EWEA (European Wind Energy Association) [1]. As a result of all the facts mentioned above, it seems absolutely necessary for us to invest on new ways of optimal energy production from wind. As it is obvious, this optimization can do either with new wind turbines models that work more efciently or with a better planning of the wind parks, in terms of wind turbine placing or location selection. Already the wind turbines are mature in technology, but for multi-megawatt production in wind farms, we have to proceed to an optimization of the production to be nancially competitive to the conventional forms of energy production. In the present study, the optimization is made by means of maximum energy production combined with the minimum cost. The basic factor that is examined is the optimal positioning of wind turbines in a wind farm, based on the criteria mentioned above. In this paper, a totally new approach has been made by using the Monte Carlo simulation method. The optimization is made through a program code that was developed in Matlab, based on the Monte Carlo simulation method.

ARTICLE IN PRESS
1456 G. Marmidis et al. / Renewable Energy 33 (2008) 14551460

2. Past approaches Many attempts have been made towards an optimal wind turbines positioning. As Bansal et al. claim in their essay, 10 ha/Mw can be taken as the land requirement of wind farms, including infrastructure [2]. Of course many conditions, like the morphology of the terrain, the speed and the direction of the wind and also the turbine size will specify the spacing between the wind turbines in a wind farm. Patel , taking account of that criteria reached, in 1999, came to the conclusion that the optimal placement for the wind turbines in a wind park is in rows of 812 rotor diameters apart in the windward direction, and of 1.53 rotor diameters apart in the crosswind direction [3]. Three years later, in 2002, Ammara et al. conclude that Patels conclusion about the sparse wind turbine positioning was inefcient, as it was not exploiting the wind energy potential of the site. Hence, Ammara et al. proposed a scheme that would yield similar production to the sparse scheme, but would have less land requirements. Ammara et al.s [4] scheme was a dense, staggered sitting scheme. Although this scheme managed successfully to reduce the land requirements, for a certain amount of wind turbines, it was still an intuitive method of placement. A completely different approach was made rst from Mosetti et al. [5] and then from Grady et al. [6] using the genetics algorithms as their basic tool in order to nd the optimal wind turbines placement in a wind farm. Mosetti et al., in 1994, were the rst to approach the optimal placement of wind turbines by means of using genetic algorithms. In Mosetti et al.s essay, algorithms were developed for wind farm performance evaluation and optimization. The whole optimization is based on nding the optimal results for the variables of investment cost and the total power extracted in the farm, which is under examination. Mosetti et al. used simple wind and cost models in order to focus more on the effectiveness of the algorithm. Because Mosetti et al.s presented congurations did not yield even the simplest empirical placement schemes. Grady et al. made a study in 2005, based again on genetic algorithms that focused on the effectiveness of the genetic algorithms optimization procedure in identifying optimum congurations. The present study is based on the same models that Mosetti et al. and Grady et al. used and we try to obtain more optimal and effective results than these two essays by using a different methodology, which is the Monte Carlo simulation method. Although we use a completely different methodology, we still use a completely similar approach, so our results are comparable. That is why the last chapter is dedicated to a confrontation between the results of the three studies. 3. Modelling The wake model used here is a simple one. Some simple assumptions were made in order to focus on the applicability of the method. Of course, it is the same model that

Mosetti et al. and Grady et al. used; so the two methodologies will be compared under the same circumstances. The model used here is similar to the wake decay model developed by Jensen [7] in 1986. Here, we assume that the momentum is conserved inside the wake, so the wind speed, u, downstream the turbine is " # 2a , (1) u u0 1 1 ax=r1 2 where u0 is the mean wind speed, x is the distance downstream of the turbine, r1 is the downstream rotor radius and a is the entrainment constant, as shown in Fig. 1. The downstream rotor radius r1 and the turbine coefcient CT are r 1a r1 rr , (2) 1 2a C T 4a1 a (3)

are related as we can see, through the rotor radius, rr, and the axial induction factor a. The entrainment constant is given empirically as a 0:5 , lnz=z0 (4)

where z is the hub height of the wind turbine and z0 is the surface roughness of the site. Every second, more than one wake is reaching the wind turbine, so it is assumed that the kinetic energy of a mixed wake is equal to the sum of the kinetic energy decits. So, the velocity downstream of N turbines can be calculated by the following expression:    n  u 2 X ui 2 1 1 (5) u0 u0 i1 or v3 u N   uX cos t u 25 . ui u0 4 1 t 1 objective Ptot u0 i1 2

(6)

u0

u rr

r1 = x+rr

x
Fig. 1. Schematic of the wake model.

ARTICLE IN PRESS
G. Marmidis et al. / Renewable Energy 33 (2008) 14551460 1457

The power produced from the wind, by a wind turbine, is a faction of the local wind speed. The characteristics that dene the local wind eld are direction, intensity and probability. Furthermore, thrust coefcient, hub height and rotor diameter also affect the power extracted. In order to calculate the total cost, we modeled the investment cost in such a way that only the number of turbines need to be considered. Mosetti et al., in order to optimize the cost, assumed that the non-dimensionalized cost/year of a single turbine is 1, and that a maximum cost reduction of 1/3 can be obtained for each turbine, if a large number of wind turbines are installed. Therefore, the total cost/year of a wind farm can be expressed as follows: cos t N2 1 e0:00174N , 3 3
2

Fig. 2. Unknown area in a known area.

(7)

where N is the total number of the wind turbines. The objective faction that will lead us to the optimal result (minimum cost per unit of energy produced) is the following expression: objective cos t , Ptot (8)

where Ptot is the total power production, while cost is calculated as mentioned in Eq. (7). 4. Monte Carlo simulation method Through the years, two basic methodologies have been developed for the analysis of systems of such huge dimensions, as we have to deal in this situation. The rst methodology, which was developed in Northern America and some European countries, is the analytical methodology. Analytical methodology is based on the enumeration of the situations that describe the whole system. The second methodology, which was developed in Italy, France and many other countries, is based on the Monte Carlo simulation method for the analysis of the system [8]. In analytical methodology, we deal only with the critical cases of the system based on the criteria we have specied in the begging of each study. On the other hand, in the Monte Carlo simulation method we have to deal with the total of the cases, a fact that it is very important in the present study, because every single positioning of the cluster of the wind turbines, which we have in our disposal, can be the optimal one. This is the reason for choosing the Monte Carlo simulation method instead of the analytical method. Monte Carlo method is a method that is based on the use of random numbers. Problems handled by Monte Carlo methods are of two types, called probabilistic or deterministic according to whether or not they are directly concerned with the behavior and outcome of random processes [9]. In this paper we have to deal with a probabilistic problem, as the solution is directly inuenced from the random numbers.

Fig. 3. Monte Carlo simulations random spots.

A common example of the Monte Carlo method is the problem of the two parallelograms. Let us, for example, use the two parallelograms shown in Fig. 2. The larger parallelogram has an area known to us, and we have to nd the area of the smallest one. This could be done using the Monte Carlo method. All we have to do is take some random spots as shown in Fig. 3. From the analog between the spots that are inside the small parallelogram and outside it, and by knowing the area of the larger one, we can have an estimation for the smaller one. Because the Monte Carlo method is an iterative method, with more iterations we can reach the result we want depending on the accuracy we specify. Of course, the larger the number of random spots, the sooner will we get to the expected result. The Monte Carlo methods application used in our project is the random placing of a certain number of wind turbines on a certain terrain. Then, by using the mathematical types and the restrictions mentioned in the previous chapter, and some computational methods, we try to reach the optimal solution. 5. Numerical procedure In this project, we have used a square grid, which was divided into 100 possible turbine locations. The total length

ARTICLE IN PRESS
1458 G. Marmidis et al. / Renewable Energy 33 (2008) 14551460

of each side corresponds to a length of 2 km. Every cell, in the center of which we can place a wind turbine, has a width that is equal to ve rotor diameters, 5D, or 200 m. So the total domain dimension is 50D 50D (2 km 2 km) as it shown in Fig. 4. The rule of the thumb spacing requirements in all directions is satised by the 5D square grid size. Furthermore, because of the width we have chosen for each cell, the wake of a column of turbines would not affect turbines in the other columns. We use a specic type of wind turbine with the characteristics shown in Table 1. The power curve for this type of wind turbine is shown in Fig. 5

The CT thrust coefcient will be considered constant throughout the processes. The power curve presented in Mosetti et al.s study for the turbine under consideration yields the following expression for power: P
N X i

0:3U 3 . i

(9)

6. Results The optimization that was made in our procedure, for a wind farm on a square-shaped terrain subdivided into 100 possible turbine locations, will be considered under a single signicant condition. The case that is under examination is the basic one: single wind direction with constant wind speed (U 12 m/s) and intensity. All calculations will start from random congurations that the program will decide, as Monte Carlo is a method that uses random numbers. In this version of the program, a better initialization will not save us any time as all cases have to be examined for the optimal result. Of course, if we introduce more parameters in our program, then a better initialization will reduce the program running time even to half or less. After examining each case, the program is made to compare the results with the previous optimal ones that were stored. If our new results from the last case are more optimal than the previous stored one, then the program automatically stores the new ones, or else it overlooks them. The program was made in order to give us nally two results, the optimal one after examining half of the cases and the very optimal one after examining all the cases. This was made in case something goes wrong during the running time of the program, so as to have some optimal results anyway. Now, there will be a comparison between the results of our study and the two previous, respective studies. The two studies we refer to are Mosetti et al.s and Grady et al.s, both based on genetic algorithms. In Fig. 6, we can see the proposed optimal placement of the wind turbines from the three studies. In Fig. 6a, we have the optimal results of Mosetti et al.s study; in Fig. 6b, the optimal results of Grady et al.s; in Fig. 6c, the optimal results of the present study, after examining half of the cases and in Fig. 6d, the optimal results after examining all the cases. All of them were congured under the same conditions. It is important to remember that the rst two studies are based on genetic algorithms and this study is based on the Monte Carlo simulation method. Table 2 is a comparison between the number of turbines, the total power output and the tness for each of the three congurations. Again as in Fig. 6, the letter (a) represents Mosetti et al.s study, (b) Grady et al.s study and (c) and (d) the two outcomes of this study. In the present study, we managed to obtaine much better production in both cases, as our total power output is 15,164 kW/year in case (c) and even larger in case (d) as it is 16,395 kW/year, compared to

U 2km

2km

Fig. 4. Subdivided wind park area.

Table 1 Wind turbine characteristics Hub height (z) (m) Rotor radius (rr) (m) Thrust coefcient (CT) 60 40 0.88

700 600 Power (kW) 500 400 300 200 100 0 0 5 10 Wind speed (m/s)
Fig. 5. Wind turbine characteristic.

15

ARTICLE IN PRESS
G. Marmidis et al. / Renewable Energy 33 (2008) 14551460 1459

x x

x x x x

x x

x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x

x x x x x

x x x x x x

x x x

x x

x x x x x x x x x x x x x x x x x x x x x x x x

x x x x x

x x x x x x x

x x

x x x

Fig. 6. Display of the results of the three studies. (a) Mosetti et al.s. (b) Grady et al.s. (c) Present study. (d) Present study.

Table 2 Comparative result table a Number of turbines Total power (kW/year) Fitness value 26 12352 0.0016197 b 30 14310 0.0015436 c 29 15164 0.0014225 d 32 16395 0.0014107

the ratio between cost and total power output. objective cos t . Ptot

the 12,352 kW/year of Mosetti et al.s and 14,310 kW/year of Grady et al.s. study It is important to notice that in case (c), even if we got this output by having more turbines than Mosetti et al.s study, we have less turbines than Grady et al.s study that is considered an optimal study compared to Mosetti et al.s one. This fact already shows that some kind of optimization has been achieve. Of course, in case (d) we have more turbines than the other three cases, which is a fact that explicate the larger power outcome, but still this does not mean that we did not succeed an optimization. As we mentioned earlier in this paper, the crucial factor that will specify the result of this optimization and whether our results are optimal or not, compared to the other two studies, is the tness. Here, we have to remind that tness is

Study (a) achieved a tness value of 0.0016197, study (b) achieved a tness value of 0.0015436 and the present study achieved an even smaller price of 0.0014225 in case (c) and the smallest of all, 0.004107, in case (d). The four results of the tness value demonstrate that the present study has reached more optimal results, in both cases, in comparison to the two previous studies. Conclusively, we managed to obtain a more optimal placement of the wind turbines as we produce much greater amounts of total power output and even greater prices of tness value. Is a certain extent, the goal of this essay has been fullled. Of course, this program has much more potentiality for even better results, which unfortunately will cost us in time. 7. Conclusions The results of the present study prove that the Monte Carlo simulation method can give us a novel approach to the tools that we already have in the eld of optimization.

ARTICLE IN PRESS
1460 G. Marmidis et al. / Renewable Energy 33 (2008) 14551460

Although we have studied and compared only one (the simpler) case, we clearly see the difference in the results between the genetic algorithms and the Monte Carlo simulation method. We managed to obtain much greater power outcomes, not always by using more turbines, and far better values of tness. Of course, this may be because of the lacking processing power or the smaller amount of probabilities that Mosetti et al. and Grady et al. decide to examine. Nevertheless, the results we have are very hopeful for the future. Still we have to examine more models and other cases to be sure. Certainly, more optimal results can be an outcome of one more complex programming code that will be the subject of a future study. Acknowledgments This work is dedicated to our mentor Professor Dimosthenes Agoris, who is no longer with us. Also in part was supported by the European Union and the Greek Government, PENED 2003, code number 03ED158.

References
[1] /http://www.ewea.orgS [accessed May 2007]. [2] Bansal RC, Shatti TS, Kothari DP. On some of the design aspects of wind energy conversion systems. Energy Convers Manage 2002;43(16): 217587. [3] Patel MR. Wind and power solar systems. Boca Raton, FL: CRC Press; 1999. [4] Ammara I, Leclerc C, Masson C. A viscous three-dimensional differential/actuator-disk method for the aerodynamic analysis of wind farms. J Sol Energy Eng 2002;124(4):34556. [5] Mosetti G, Poloni C, Diviacco B. Optimization of wind turbine positioning in large wind farms by means of a genetic algorithm. J Wind Eng Ind Aerodyn 1994;51(1):10516. [6] Grady SA, Hussaini MY, Abdullah MM. Placement of wind turbines using genetic algorithms. Renew Energy (Elsevier) 2005;30:25970. [7] Jensen NO. A note of wind generator interaction. Roskilde, Denmark: Risc National Laboratory; 1993. [8] Dialynas EN. Simulation Monte-Carlo. Athens: October 1996. [9] Hammersley JM, Handscomb DC. Monte Carlo methods. London: Chapman & Hall; 1979.

You might also like