Context
Academic — Technische Universität München, Design of Wind Farms
Role
Course project · layout & BOS lead
Stack
NREL FLORIS, NREL LandBOSSE, Python (NumPy / Pandas / Matplotlib)

Problem

A candidate wind-farm site had to be evaluated end-to-end: turbine selection, micro-siting within a bounded lease area, wake-loss-aware power estimation, and balance-of-system (BOS) cost — which for real farms is half the capital stack, not an afterthought. The challenge was that the optimisation trades were non-obvious: a layout with lower wake loss might require longer inter-array cabling or more expensive foundations, and nothing in the problem statement told you the winner a priori.

Approach

  • Site definition — defined the project boundary as a quadrilateral polygon with four corner coordinates (WGS84) covering ~9 km² of agricultural land in southern Jutland, Denmark.
  • Constraint screening — applied legal buffers (740m residential setback, 30m road buffer, 100m power line buffer, 300m forest buffer), ecological buffers (30-100m from wetlands/meadows/streams, 50m voluntary wildlife buffer), and logistical buffers (60m from field boundaries for crane access).
  • Buildable area — intersected the site boundary with all constraint buffers to identify the final usable area for turbine placement. Less than one-fifth of the total site remained after all exclusions.
  • Resource assessment — built a site wind rose from reanalysis data; fit Weibull distributions per sector. Dominant wind directions: North (0°) and West (270°), ~29% of total occurrences, with scale parameters 9-11.7 m/s.
  • Turbine selection — selected IEA 3.4 MW reference turbine (130m rotor, 120m hub height) based on specific power (256 W/m²), logistics (65m blades transportable on Danish roads), and cost efficiency (16-25% cheaper per MW than larger options).
  • Wake modelling in FLORIS — NREL's FLORIS framework with Gaussian wake model. Swept turbine counts (14-18 turbines) for 45-60 MW target capacity.
  • Layout optimisation — FLORIS's layout-optimisation routines within the lease boundary with minimum-spacing constraints (5D prevailing, 3D cross-wind). Compared structured grid, staggered, and FLORIS-optimised free layouts.
  • BOS costing in LandBOSSE — NREL's LandBOSSE model for civil works, erection, collection-system cabling, and grid interconnection. Each candidate layout costed against the same turbine and site conditions.
  • LCOE close — AEP (from FLORIS with wake losses and availability) divided through lifetime cost (LandBOSSE + assumed O&M and turbine CapEx) to give a defensible LCOE ranking.
Site 1 boundary polygon — 9 km² agricultural land in southern Jutland, Denmark
Site definition — quadrilateral boundary polygon (Pt1-Pt4) covering ~9 km² near Kastrup, Denmark. Coordinates: Pt1 (55°15'48.06"N, 9°0'48.68"E) to Pt4 (55°14'39.19"N, 9°0'41.31"E).
Legal, ecological, and logistical constraints for wind farm siting
Siting constraints — legal (740m residential, 30m road, 100m power line, 300m forest), ecological (30-100m wetlands/meadows, 50m wildlife), logistical (60m field boundaries for crane access).
Buildable area after constraint screening — final usable area in dark green
Buildable area — constraint screening reduces the site to less than one-fifth of total area. Light mint: initial buildable. Light blue: buffered field boundaries. Dark green: final usable area.
Wind rose for Site 1 Denmark showing prevailing wind directions
Wind resource — Site 1 wind rose showing dominant North (0°) and West (270°) directions. Weibull scale parameters 9-11.7 m/s confirm excellent wind resource potential.
Initial vs optimized turbine layout — red circles initial, blue X optimized, green square substation
Layout optimization — red circles: initial layout. Blue X: FLORIS-optimized layout. Green square: substation. Optimized layout reduces wake losses through staggered positioning.
Wake flow comparison — initial aligned layout vs optimized staggered layout at hub height
Wake comparison — initial aligned layout (left) shows significant wake overlap in downstream columns. Optimized staggered layout (right) positions downstream turbines in wake gaps, reducing speed deficit.
Wake flow comparison for 18-turbine layout — initial clustered vs optimized spaced arrangement
18-turbine wake comparison — initial clustered layout (left) with heavy wake overlap (black ellipses). Optimized layout (right) increases spacing, allowing wake recovery before downstream turbines.

Tools & setup

  • FLORIS (NREL) — engineering wake models, layout and yaw optimisation; the industry reference for wake-loss-aware plant design.
  • LandBOSSE (NREL) — balance-of-system cost model spanning civil, erection, collection, management, and grid interconnection.
  • Python — driver scripts stitching resource → FLORIS → LandBOSSE → LCOE in one pipeline, with pandas DataFrames passing results between stages.
  • Matplotlib — wake-field plots, AEP-by-sector charts, and the final cost-vs-AEP scatter used for layout selection.

Outcomes

  • Turbine choice justified on the data — the chosen rotor diameter and hub height maximised AEP per installed MW given the site's Weibull scale and turbulence class; swapping to a larger rotor was rejected on BOS (foundation and crane) grounds even though it improved wake-free AEP.
  • Wake-aware layout — the staggered, prevailing-wind-aligned layout beat the simple grid by a meaningful AEP margin while only slightly increasing collection cabling, so it won on LCOE.
  • Ranking by LCOE, not by wake-loss alone — the defining methodological choice. Two layouts with similar wake losses had materially different BOS costs, and that is what the ranking surfaced.

Headline takeaway: the "best" layout from a pure wake-loss perspective was not the best on LCOE. Cable length and crane logistics flipped the ranking. This is the lesson LandBOSSE is designed to teach — BOS is where good wake engineering either earns its keep or doesn't.

Validation

FLORIS predictions were cross-checked against the simpler Jensen model as a floor — the Gaussian wake model gave slightly higher partial-wake recovery than Jensen, consistent with the published literature. Resource AEP numbers were sanity-checked against a capacity-factor hand-calc from the site Weibull parameters and the turbine power curve; the two approaches agreed to within the expected wake-loss gap. LandBOSSE outputs were compared against published onshore farm BOS ratios (EUR/MW installed) and fell in the expected band.

Lessons learned

  • Wake modelling is the headline physics, but BOS is where layout decisions are won or lost. Decoupling the two leads to wrong answers.
  • Engineering wake models (FLORIS-class) are the right tool for layout sweeps — the per-evaluation cost is low enough to run thousands of candidates, which CFD cannot do.
  • Turbine selection should be part of the same optimisation, not a precursor to it. A slightly smaller rotor with cheaper foundations sometimes wins.

Artifacts

  • Design of Wind Farms course report — methodology, results, and layout comparisons.
  • FLORIS + LandBOSSE driver scripts (sanitised, available on request).