Wind Farm Energy Assessment

Case study with figures: zaydhouachmi.github.io/work/wind-farm.html · Try the wake model in your browser

A Streamlit application that estimates the annual energy production of a wind farm from wind measurements, terrain roughness and a turbine layout. Built during an internship at Greenard as a lightweight, open alternative to commercial tools such as WAsP.

What it does

1. Wind resource — builds a wind rose from a dominant direction, a manual 12-sector table, or an uploaded CSV (either sector statistics or raw measurements, with Weibull parameters fitted by the method of moments).
2. Vertical extrapolation — scales wind speed to hub height using the logarithmic profile and the site roughness length z0.
3. Wake modelling — computes velocity deficits between turbines (Jensen and Ainslie-derived models) so downstream turbines see the reduced wind.
4. Energy yield — integrates the power curve over the wind distribution, then applies a cascade of six loss factors (shadow flicker, flow effect, availability, electrical, turbine performance, environmental).
5. Results — gross / after-wake / net energy, capacity factor, and total wake loss, with layout and wind rose plots.

Results

Fed the 12-sector Weibull table from the WAsP "Waspdale" example site, the tool recomputes a 4.90 m/s mean wind speed, the same value WAsP reports. A 6-turbine, 1 MW farm (54 m rotor, 50 m hub, 5 rotor diameters apart, Jensen wakes) then shows why height and terrain correction matter:

Wind climate	Mean speed	Gross energy	Net energy	Capacity factor
Raw station data (10 m)	4.90 m/s	3,679 MWh/yr	2,663 MWh/yr	5.1 %
Corrected to the turbine site (50 m)	7.88 m/s	16,186 MWh/yr	9,264 MWh/yr	17.6 %

The second row puts all the wind in one direction (270°), which is the worst case for wakes: up to 62 % loss on the third turbine in a row. A real wind rose spreads the wakes out. tests/test_model.py checks that behaviour.

Getting started

pip install -r requirements.txt
streamlit run app.py

Then open the URL Streamlit prints (usually http://localhost:8501).

Input file formats

Example CSVs are included so you can see the expected columns:

File	Columns
wind_rose_example.csv	direction_deg, frequency_pct, weibull_k, weibull_a
wind_measurements_example.csv	direction_deg, speed_ms (one row per reading)
power_curve_example.csv	wind_speed_ms, power_kw
layout_example.csv	id, x, y (metres)
roughness_example.csv	z0

Project layout

Module	Role
app.py	Streamlit interface — all inputs and results
wind_resource.py	Wind rose, Weibull fitting, vertical extrapolation
power_curve.py	Parametric and tabulated turbine power curves
wake_model.py	Wake deficit models
energy_yield.py	Gross/net energy and the six-factor loss cascade
turbine.py, wind_farm.py	Turbine and farm data structures
topography.py	Terrain roughness handling
visualization.py	Layout and wind rose plots
demo_*.py	Runnable examples that produce the images above
tests/	pytest suite for the model

PHASE_2_AINSLIE_DERIVATION.md documents the maths behind the wake model, and CHANGELOG-v3.md records what changed between versions (in French).

Tests

pip install pytest
python -m pytest

22 tests in tests/test_model.py check the model against hand calculations: the Jensen wake formula, the power curve, Weibull statistics, log-law height scaling, the loss cascade, wind-aligned geometry, and the farm energy table. demo_smoke.py is a quick script that compares the three wake models on one layout.

Known limitations

• No orographic correction for terrain relief — only roughness and height are accounted for.
• The thrust coefficient Ct is fixed (0.8 by default) rather than varying with wind speed.
• Weibull fitting from raw measurements uses the Justus method of moments, not maximum likelihood — good enough for estimation, less precise than a dedicated tool.