Including Wind Resource Trends in a Micrositing Optimization of the Block Island Wind Farm, RI, USA

The exponential growth of the earth’s population has lead to the depletion of natural resources in concert with unrepairable environmental damage. One solution for a more sustainable lifestyle is the supply of electricity by renewable energy technology. Offshore wind energy is expected to play a major role in the extension of this sustainable energy supply. However, several challenges lay ahead due to the high expenses of offshore energy. Consequently, the optimization of a wind farm layout for minimizing costs and maximizing revenue gains high importance. This study determines the sensitivity of a wind farm layout and its revenue to wind time series length, wind direction and wind velocity trend. The sensitivity analysis is conducted at the Renewable Energy Zone (REZ) and the Rhode Island and Massachusetts Area of Mutual Interest (AMI) in Rhode Island, USA. Optimum layouts are found by minimizing an objective function expressed in terms of the Wind Farm Siting Index (WiFSI) with a Genetic Algorithm. The objective function considers wind resource, technological costs as expenses for tower foundation and cable interconnection as well as ecological and fishery cost. Ecological cost is expressed as abundance and sensitivity of species to wind farms. Fishery cost is implemented proportional to fishing activity intensity. The WiFSI is a dynamic tool adjustable by weighting factors to societal or political choices. Seven simulations are conducted for the REZ and four simulations are researched for the AMI to complete the sensitivity study. All scenarios exclude ecological and fishery constraints from the objective function to focus on effects of a changing wind resource. Simulation 1 is conducted as a base case for the REZ with constant wind resource measured over three years. Simulation 2 to 4 apply wind probability distributions of 1992 to 2012, 1980 to 2012 and 2008 to 2012, respectively. Applying the long wind time series leads to several placement solutions in contrast to one optimum layout for simulation 1. Scenario 5 and 6 apply single-year wind roses of the years 1980 and 2012. Resulting layouts differ in orientation to the respective dominant wind direction. Simulation 7 implements a positive, linear trend in usable wind velocity. The same layout as for simulation 1 is found but net revenue increases. One base case simulation with constant wind and one simulation with a positive, linear trend in the usable wind resource are operated for the Northern and the Southern part of the AMI. Optimized layouts of the base case and the trend application simulations vary. The change is due to the high number of possible turbine locations with same WiFSI. The implementation of the trend leads to an increase of net revenue. Wind turbine layout is highly sensitive to a change in wind time series length and wind direction while net revenue is less influenced. In contrast, the sensitivity of the layout to a trend in usable wind velocity is low while the effect on net revenue is significant. Conclusively, long-term wind predictions over the life time of a farm are necessary to determine the optimum layout as well as produced power of a site.

Due to the depletion of natural energy resources offshore wind energy has become of great interest for sustainable electricity supply. Regarding the high expenses for construction and maintenance of a offshore wind farm, research focuses on finding a turbine layout with minimum cost and maximal power production.
This study overcomes the limitation of using a constant wind resource by implementing changing wind time series lengths, varying wind direction and a trend in usable wind velocity. The sensitivity study is applied to the Renewable Energy Area (REZ) with five turbines as well as the Rhode Island and Massachusetts Area of Mutual Interest (AMI) with 100 turbines. The changing wind resource is implemented in the methodology of the Wind Farm Siting Index (WiFSI). This tool combines technological, ecological and fishery constraints in an objective function to find optimum turbine locations. Weighting factors for each constraint are adjustable to societal and political choices. Wind turbine placement is determined to be highly sensitive to a change in wind time series length as well as changing wind direction. In contrast, the impact of a trend in usable wind speed on a layout is low while net revenue is changing significantly. Conclusively, long-term wind predictions over the life time of a farm are necessary to determine the optimum layout as well as produced power of a site.

Keywords
Offshore renewable energy, Wind farm micrositing, Wind farm siting index, Wind trend, Ecosystem services, Genetic algorithm

Introduction
The exponential growth of the earth's population has lead to the depletion of natural resources in concert with unrepairable environmental damage. One solu-tion for a more sustainable future is the supply of electricity by renewable energy technology. For several countries, targets range up to 100% of energy supply from renewable sources [1]. As of 2012, 19% of the world-wide energy consumption was produced by renewable energy plants. Offshore wind energy is expected to play a major role in the targeted argumentation of sustainable energy supply. Providing emission-free electricity from a domestic, non-ending source, this technology also strengthens the economy and diversifies the energy supply leading to independence from conventional, environmental unfriendly energy devices [2] [3]. However, several challenges lay ahead since offshore energy requires high expenses due to the foundation, installation and maintenance at the open sea [4]. The optimization of a wind farm layout gains high importance in this context [5]. Consequently, recent efforts to site wind farms have resulted in placing a given number of turbines in a predefined area based on a siting optimzation approach which aims on maximizing the revenue [6].

Wind farm siting in Rhode Island
The state of Rhode Island, USA, passed the Rhode Island Energy Standard (RES) in June 2004 in order to tackle the challenges of offshore wind energy. This portfolio demands that by 2019 16% of the total supplied electricity must originate from renewable sources [7]. One contribution to this goal is the first US offshore wind farm Southeast of Block Island which is currently under construction [8]. The location of the Block Island Wind Farm was selected based on an extensive siting optimization research phase performed through the Ocean Special Area Management Plan (Ocean SAMP) [12]. This eco-system based coastal management tool was initiated by the Rhode Island Coastal Resources Management Council (RI CRMC) and developed in close collaboration with the University of Rhode Island.
The study area is shown in figure 1. The SAMP area is home to migratory fish, sea Figure 1. Map of the site of the Special Area Management Plan (SAMP) (outlined in red) with water depth in meter (color shading) from Grilli et al. (2014) [9]. turtles, birds and marine mammals in addition to the presence of several human activities such as ship traffic or recreational and commercial fisheries [10]. These ecological and human activities constitute significant constraints to wind farm development which were tackled in the SAMP project through scientific research and stake-holder meetings. In that context,  [11] developed a siting optimization protocol including ecosystem service constraints in a wind farm cost model that improved the classic wind farm optimization. Based on the comprehensive Ocean SAMP study [12] and its associated siting optimization tools, a small offshore Renewable Energy Zone (REZ) (Figure 2) was defined in state waters as well as a much larger zone in federal waters. The large zone is called Area of Mutual Interest (AMI) and expands across the offshore area of Rhode Island and Massachusetts ( Figure 3) [13] [14].   [11].

Classic wind farm siting optimization
Conventionally, two approaches are followed in research on wind farm siting optimization. The first approach focuses on the development of siting algorithms applied to simple test cases [15] [16] [17], while the second approach performed in studies such as TopFarm [18] or OWFLO [19] focuses on the refinement of the aerodynamic modelling. These studies validate their model with measurements at existing wind farms. Therefore, the difference in the approaches lies in the complexity of the objective function, the formulation of the constraints to be minimized, and the sophistication of the aerodynamic and economic models. Table 1 gives an overview of previously applied objective functions in both approaches. Wan et al. (2010) [17] rely on wake effects as single constraint in Annual   [20] and Tesauro et al. (2010) [6] state that maximizing energy production alone is far from sufficient and that all costs of the farm must be included in the cost model to achieve a realistic optimization. Grady et al. (2005) [16] define Cost Of Energy per KWh (COE) as ratio between total investment cost and total extracted power to test a Genetic Algorithm. Mosetti et al. (1994) [15] introduce the rarely used concept at that time of a linear combination of costs where weighting factors can be adjusted to economic choices. Adding expenses for operation and maintenance as further constraints in the model leads to the Levelized Production Cost (LPC) applied by Elkinton et al. (2006) [19] or  [21]. Réthoré et al. (2014) [22] define a profit function as Financial Balance (FB) including installation cost, cable cost, fatigue degradation cost and discounting rate while Gonzáles et al. (2009) [23] use the concept of Net Present Value (NPV) of an investment for the optimization process.

Current research in wind farm siting
Currently, the classic siting tools' critical components are the wake modelling [21][19] [27] and a precise cost representation [6]. Barthelmie et al. (2006b) [28] compare measurements and predictions of the most commonly used wake models. None of these is able to precisely predict wind speed in the nearfield behind a turbine. Gonzáles et al. (2014) [4] therefore suggest a revaluation of the current wake formulations. Elkinton et al. (2006) [19] miss precise cost models for offshore wind farm availability, installation cost as well as expenses for operation and maintenance while Réthoré et al. (2014) [22] demand refinement of the electrical grid models. Improvement of component reliability analysis is requested by Gonzáles et al. (2014) [4].
Restrictions of the conventional models has lead to the development of new approaches to wind farm siting. In particular, the standard computational grid in optimization models has evolved towards a continuous siting approach. Wan et al. (2009) [24] first allowed the devices to be placed freely inside a computational grid cell instead of its center to further reduce the wake. In parallel, a new concept of flexibility in terms of turbine parameters has emerged. Acero et al.
(2009) [29] apply variable hub heights for turbines aligned on a straight line while Chen et al. (2013) [30] keep number of turbines and height as free variables when optimizing a simple test site. The Unrestricted Wind Farm Layout Optimization (UWFLO) developed by Chowdhury et al. (2013) [31] as well as Rahbari et al. (2014) [32] combine several of these approaches. The model optimizes under continuous placement conditions with variable rotor diameter, variable hub height and variable power characteristics.

Social and environmental constraints
Studies on the impact of wind farms regarding social and environmental factors have been conducted in the last years followed by the development of models to predict specific impacts of wind farms on the environment. Visual impact increases with number and size of the turbines [33] and decreases with distance to shore [34].
Visual impact is also dependent on arrangement, spacing, color and material [35] as well as on personal attitude towards offshore farms [36]. Shadow impact decreases with distance and flicker frequency while offshore noise level depends on wind direction and distance [37]. A noise emission model for wind turbines was developed by  [38]. Bird collision risk at wind farms is lower than with other man-made infrastructures [39]. Nevertheless, even low additional mortality may be significant for slow reproducing species [40]. Band et al. (2007) [41] developed a collision risk model based on flight height, avoidance behaviour and turbine type. Overall, the most effective mitigation technique seems to be shut-down of turbines on demand [42]. Constructional or operational noise is expected to result in displacement or a shift in common migration routes of mammals [43] while fish abundance is larger in the presence of turbines (reef effect) [44][45].

Limitation of current siting tools
Examining the factors taken into consideration in the objective functions of the conventional micrositing tools it becomes obvious that social and environmental factors are missing in most algorithms although several frameworks and models regarding these impacts have been developed in the past years [11]. El-Thalji et al. (2011)[20] introduce the implementation of cold climate as a factor of wind farm siting. Staffel and Green (2014) [46] record turbulent loads and determine output declination. Chowdhury et al. (2013) [31] propose the implementation of land area per kW installed as further decision making tool. Christie and Bradley (2012) [47] maximize power density as power per unit area of land occupied to overcome limited available sites.
Still, these first approaches to include environmental issues in micrositing tools aim on maximizing energy production. Few studies involve options for minimizing potential environmental impacts of a wind farm. Kwong et al. (2012) [48] run a Genetic Algorithm for optimizing a simple test case while taking in account energy production and noise impact. Overall, there is a lack of research regarding implementation of ecological and social constraints in micrositing tools for the ecological sensitive marine area [49].
Another factor missing in current siting tools is the change in the wind resource over time. Wind speed is the primary factor to a wind farm project [6] since produced power is directly proportional to the cube of wind speed. Satellite altimeter measurements revealed an increase of wind velocity of at least 0.25 to 0.5% over the last 20 years with speeds of extreme events increasing faster than mean conditions [51]. Wind speeds in Spring rise faster than in Fall and meridional wind trends are smaller than zonal trends [52]. Although many studies show a positive trend over the last decades, the magnitudes vary significantly and require further investigation [53].
Conventional siting tools used to apply several wind models to research the effect of wind conditions: unidirectional uniform wind, uniform wind with variable direction and non-uniform wind variable direction [16][17] [54]. The complex models either use discrete wind distributions [16] [15] or continuous Weibull distributions [31] [55][32] [56]. To the best knowledge of the authors, no model con-siders the previous described change in the wind resource over the life time of a wind farm.

Study objective
The current study researches the effects of changes in the wind resource over Interest.

Materials and Methods
The  [63], the optimum location of each turbine is determined using a Genetic Algorithm to optimize an objective function (OF). The objective function is derived from the Wind Farm Siting Index (WiFSI) which was developed as a tool to optimize macrositing of a wind farm by  [11].

The objective function
The WiFSI describes a non-dimensional balance between technological, social and ecological constraints and the wind resource [11]. The conceptual view of the index is given in figure 4. Each constraint is weighted with a factor w i which is adjustable according to societal values or political choices [62]. Social Figure 4. Concept of the Wind Farm Siting Index (WiFSI) as balance between constraints and resources from  [11]. The constraints are technological cost (TC) as well as ecosystem services cost (ESC) divided into ecological and fisheries service. The resource is wind power (PP).
and environmental value are formulated as the services which the ecosystem provides to human beings [61]. Since these intangible costs [59][60] can not directly be described in monetary terms [58] where the net revenue (PR) is based on a monetary balance between technological costs and extracted power. The scaling factor (SF) is equal to the worst possible income at that site. This factor non-dimensionalizes the net revenue to be comparable to the non-monetary ecosystem service constraints. EC and FC are mean ecological and fishery costs for the respective turbine locations.
The net revenue of n-placed turbines is the difference between the produced power in $/kWh including extractable power (P ex ) and power loss due to wake (PL) as well as technological cost (TC) [63]. The extractable power is defined as the usable power density (P u ) in W/m 2 at the rotor level restricted by Betz Law (P ex = 59.4%P u ). In the concept of usable power, wind speeds lower than cut-in as well as higher than cut-out velocity are ignored while velocities higher than rated wind speed are kept constant for the power calculation [64]. The power law is applied to calculate wind speed at rotor elevation [11]. Power loss is calculated with the WAsP Model formulated by  [65]: where U loss is the velocity deficit behind the turbine and U f reestream the arriving wind velocity. C T is the thrust coefficient of the turbine. The rotor diameter is expressed by D and the wake diameter D w describes the lateral extent of the wake in a distance x from the turbine. The model is based on a simple one-dimensional concept where the momentum deficit is expanding linearly behind a turbine with a wake decay coefficient k of 0.05 for offshore sites. Wind speed data is discretized into 1 degree sectors. The data is then expressed in mean wind speed and probability of occurrence for each directional sector [63].
Technological costs represent the technical challenge resulting from the placement of one turbine at a given location. Expenses for foundation and installation as a function of water depth and geological characteristics as well as costs for cable connection depending on distance are considered. Non-relevant costs for micrositing such as expenses for maintenance or device retail price are excluded through the relative cost principle [11] [22].
Ecological costs are expressed in non-monetary value based on seasonally averaged abundance of fish species and mammals as well as their intrinsic sensitivity to noise, turbidity and electromagnetic field perceived during the construction and the operation phases. Fishery costs are included as a linear combination of commercial mobile and fixed gear fisheries as well as recreational fishing activities. The costs are assumed proportional to fishing intensity [11].  [63].
are [63] [66]: The model developed in Matlab is amenable to parallel processing and uses the Genetic Algorithm Toolbox [67]. The siting area is gridded with one rotor diameter discretization. Turbine placing is restricted to the center of a grid cell to increase rate of convergence [63]. The distance optimization algorithm for the cable interconnection is a Cluster Algorithm at the REZ. Due to increased turbine number

Implementation of changing wind resource
The trend in wind velocity is determined by linear regression in a 95% confidence interval over the hourly usable wind speed values. The resulting slope is applied as yearly trend in the calculation of extractable power and power loss due to wake. The sensitivity of a change in wind direction on the optimization layout is tested by implementing different usable wind roses while keeping the power resource constant.  Standard wind rose, trend in speed 10 AMI North 60 Alstom 6MW Standard with constant wind 11 Standard wind rose, trend in speed In order to create long-term usable wind roses, wind data is taken from the WIS Station 63101 and cut for the usable wind velocity range. For simulation 2  corresponds to the standard length of a representative wind time series [50].
The usable wind rose from 1992 to 2012 can be seen in figure 8.   Therefore, the objective function reduces to:   Algorithm was validated by application to the REZ. Computational time is reduced by excluding the trend in the usable wind velocity from the wake calculation.
Testing this method at the REZ leads to a 0.02% loss of produced power. This simplification was considered acceptable and implemented in the AMI zone in order to keep the computational time in a reasonable time frame (days versus weeks).

Results and Discussion 1.5.1 Renewable Energy Zone Base case
Optimum solutions were found for 5 distinct locations. One solution is shown in figure 12. The cable interconnection costs pull the turbines together while the wake effect pushes the devices apart. The turbines are placed in areas of low foundation cost. Extractable power has no significant influence since the resource is approximately equally distributed over the zone [63]. Figure 13 shows that all locations are sited in the area of the lowest WiFSI of the REZ. The East part of the REZ with a higher index is avoided due to high foundation cost. The map was created for macrositing purposes in  [11] with the initial formulation of the WiFSI. Compared to the layout which was developed by Deepwater Wind, the optimized solution would save $17 Million over a lifetime of 20 years due to the reduction of wake effects [63].

Sensitivity to time series length
The first runs were operated with 1000 generations to test convergence for    The determination of several solutions to one optimization problem offers various options for soft constraints. The user is able to choose between layouts based on societal or political preferences. Although the optimization might exclude ecological and fishery cost, a layout with low EI or FI could be chosen or visual concerns could be solved without significant losses in total revenue.
The decrease in revenue using the long-term usable wind roses is misleading.
Total revenue is calculated assuming that the wind conditions described by the  (Table 7).
Assuming the layout optimized for the usable wind rose of 1980 but applying the long-term usable wind rose from 1980 to 2012 results in a loss of produced power and total revenue of $0.86 million or 0.33% over 20 years. When the usable wind   Different wind roses result in different layouts. The more significant the variation in the wind rose, the more noticeable is the difference in the turbine locations.
Assuming the same wind conditions as the farm was optimized for, different layouts can produce the same revenue. However, the application of real life wind conditions with changing wind direction leads to reduced revenue over time. The determination and consideration of future changes of wind direction in an optimization process gains importance.

Trend in usable wind velocity
The turbine locations converge towards the same locations as the one obtained for the base case in simulation 1 (constant wind speed) ( Figure 21). The averaged installation costs remain constant but mean produced power increases by 2.17%.
Total revenue increases to $209.07 million compared to $203.55 million in the base case.  The application of a positive trend to the usable wind velocity results in an increase in power production while the turbine locations remain quasi identical.
A higher trend magnitude might have a larger effect on the layout since the wake effect will increase with higher wind velocity. However, this effect may not play a large role for wind farm siting since measured wind trends have been found considerably small [51][52] [53].

Area of Mutual Interest
Base case at the South Lease    Figure 24. Optimized layouts in simulation 8 at the South Lease (grey) applying the usable wind rose at WIS Station 63095 from 2008 to 2012. Each color-set represents one optimum layout. North Lease Trend 6.05 1.15 7.20 the optimization of the REZ. All runs end in the same range of produced power and installation cost resulting in an average total revenue of $3.95 billion (Table 9).

Base case at the North Lease
The dependence of the WiFSOF to number of generation when optimizing the North Lease shows the typical shape for convergence.  Figure 25. Convergence of layouts in simulation 10 at the North Lease (grey) applying the usable wind rose at WIS Station 63095 from 2008 to 2012: after 1400 generations (red), 900 (blue), 400 (light blue), 2 (green).
an additional revenue of $88.85 million or 1.49%. After 1000 generations only four turbines change location. This is a difference in total revenue of $2.06 million or 0.004%. Figure 25 shows that the higher the generation number, the more likely the turbines align at the borders of the lease. The small allowed area in the Northwest is avoided as well as the Northwest border. Only three rows of nine devices do not follow a border. Net revenue is $5.99 billion for the final layout (Table 9).

Trend in usable wind velocity at the South Lease
Applying the trend in usable wind velocity leads to a similar layout as the one   Figure 27. Optimized layouts in simulation 10 at the North Lease (grey) applying the usable wind rose at WIS Station 63095 from 2008 to 2012: constant wind (blue) and implementation of a usable wind speed trend (red).
central area is mainly avoided. Four rows of 14 turbines are placed across the area.
The angles of the rows vary. Total revenue is $6.05 billion which is an increase of $56.87 million or 0.95%.

Discussion
The optimized layouts for the base case simulations are not as similar to the layouts of the simulations with the trend in usable wind speed as in the REZ. This   [63] showed that turbines tend to spread over the whole siting area when interconnection cost are not included. In that case, the power lost due to wake effects decreases. In contrast, turbines cluster at one point at a specific site when wake is excluded but interconnection is the major constraint. At the North Lease, the cable costs constraint leads to small distances between each turbine and its neighbours by placing them in rows. The wake effect is expressed through distance maximization between each row. The distance between the turbines is restricted by the zone borders and the number of turbines.
Both AMI zones were optimized separately to reduce computational time. The layout of the leases changes when optimized together. Areas of high foundation cost would still be avoided but interconnection cost and wake effect would change the overall shape of the turbine layout.

Comparison
Installing the turbines of the optimized layouts of the North and South Lease would cost $2.09 billion and produce power for $12.05 billion. Assuming the trend in usable wind velocity, net revenue increases from $9.94 to $10.04 billion. This increase is 1.0% or $94.42 million. Average installation cost per turbine is less for the North Lease due to lower water depth ( Figure 31). The South Lease outweighs higher installation cost with a higher power production per turbine because of higher wind velocity ( Figure 30). Optimizing the REZ with wind data from a different station as the AMI but the same time period from 2008 to 2012 leads to 60% less total revenue per turbine.
Installation cost is cut in half but produced power is only 42% of the power of the AMI. It has to be considered that different types of turbines have been used for the REZ and AMI. On the one hand, the Siemens 6MW used at the REZ is 5 m taller and rotor diameter is 4 m longer than the Alstom Haliade 6MW used at the AMI. The range of usable wind speed of the Alstom device is also 1.5 m/s lower.
In addition, its rated wind speed is 2.5 m/s lower. On the other hand, efficiency of the Alstom turbine is higher (Figure 2). The difference in efficiency as well as the higher extractable power at the AMI result in the high produced power per turbine at the North and South Lease.  For all results it has to be considered that the applied usable wind rose only effects the wake calculation. The wind power resource at the sites is constant for all simulations. Only produced power is changing according to the wake loss due to the relative positions of the turbines. In addition, the applied usable wind rose is assumed to be representative for the entire life time of a farm.

Conclusions
This study determined the sensitivity of a wind farm layout and its total revenue to wind time series length, wind direction and usable wind velocity trend. and wind direction is significant. The impact on the total revenue is smaller but recognizable. This is because all other constraints as foundation cost and extractable power remain the same for all simulations while only the wind direction changes. In contrast to that, the implementation of a trend in usable wind velocity does not result in a significant change in the turbine layout but has a a higher impact on the total revenue. A positive trend leads to an increase in produced power while installation cost remain the same. Total revenue rises.
A long-term wind prediction over the life time of a farm is necessary to determine the optimum layout and expected total revenue of a site. Short-term wind probability distributions are not representative for future wind conditions.
Their application does not result in the optimum layout for a wind farm over its entire life time. Real life income might be less than predicted total revenue. In contrast to that, long-term predictions include changes in the wind resource i.e. direction and velocity. Applying changes in the wind direction leads to wind farm designs which minimizes the impacts of the changing wake effect expansion. In contrast to that, considering the trend in usable wind velocity is not significant for finding optimum turbine locations as long as the trend is small. High trends increase the extent of the wake effect and are expected to change the layout. The implementation of the trend is therefore only necessary in economic models when expected produced power can be calculated more accurately. Economic decision can be taken regarding the decrease or increase of total revenue due to a negative or positive trend respectively.
Future work should focus on the validation of the wind farm siting model.
Model results could be compared to data collected at the Block Island Wind Farm which is currently under construction. Calculated produced power is validated with values of power produced at the Block Island Wind Farm. The one-dimensional WAsP Model can be compared to measurements of the wake decay behind the operated turbines. The influence of changing wind direction as well as speed could be observed with data of the produced power for short as well as long time periods.
Further research is required for the optimization of the AMI. More runs where constraints such as wake effect or interconnection costs are excluded are necessary to determine their sensitivity to the layout. The optimum number of turbines at the AMI could be found to maximize total revenue. Ecological as well as fishery costs could be included in an optimization. The resulting layouts show options for the usage of the marine resources when environmental, economic and social impact are considered.