WFCRL: Interfacing and Benchmark Reinforcement Learning for Wind Farm Control

Environments

List all environments with:

from wfcrl import environments as envs
envs.list_envs()

All wind farms environments are implemented with both the Gymnasium and PettingZoo API, and can be run on both the Floris and the FastFarm wind farm simulators.

The root name of the environment is associated with a specific layout, of arrangement of turbines in the field. It is combined with a prefix and a suffix:

• A Dec_ prefix is added before environment names to indicate an Agent Environment Cycle implementation supported by PettingZoo.
• A Floris or Fastfarm suffix is added after the name of the environment to indicate the name of the background simulator.

Root Name	# Agents	Description
Ablaincourt	7	Inspired by layout of the Ablaincourt farm (Duc et al, 2019)
Turb16_TCRWP	16	Layout of the Total Control Reference Wind Power Plant (TC RWP) (the first 16 turbines)
Turb6_Row2	6	Custom case - 2 rows of 3 turbines
Turb16_Row5	16	Layout of the first 32 turbines in the the CL-Windcon project as implemented in WFSim
Turb32_Row5	32	Layout of the farm used in the
TurbX_Row1 for X in [1, 12]	X	Procedurally generated single row layout with X turbines,
Ormonde	31	Layout of the Ormonde Offshore Wind Farm
WMR	36	Layout of the Westermost Rough Offshore Wind Farm
HornsRev1	76	Layout of the Horns Rev 1 Offshore Wind Farm
HornsRev2	92	Layout of the Horns Rev 2 Offshore Wind Farm

A visual overview of some layouts:

Turb7_Row1	Ormonde	HornsRev2

Example

Creating a wind farm environment of the Ablaincourt layout with the Floris background on Gymnasium:

from wfcrl import environments as envs
env = envs.make("Ablaincourt_Floris")

Examples of test cases using the PettingZoo environment are given in the examples folder. To simulate the Ablaincourt layout on FLORIS:

python examples/example_floris.py

More detailed examples can be found in the demo.ipynb notebook. See below under Running Example Notebooks.

Installation

In the virtual environment of your choice:

pip install -e .

To get the environments simulated with FAST.Farm:

• On UNIX

wfcrl-simulator fastfarm

This requires you to have cmake and an MPI implementation like OpenMPI 3.

• On Windows

1. Download the compiled FAST.Farm 3.5.1.binaries here.

2. Download BOTH Windows MPI setup (.exe) and MPI SDK (.msi) and install them from (https://www.microsoft.com/en-us/download/details.aspx?id=100593) You can check your installation by enBtering : set MSMPI from C:\Windows\System32 in the command prompt. You should obtain the following:

MSMPI_BENCHMARKS=C:\Program Files\Microsoft MPI\Benchmarks\
MSMPI_BIN=C:\Program Files\Microsoft MPI\Bin\
MSMPI_INC=C:\Program Files (x86)\Microsoft SDKs\MPI\Include\
MSMPI_LIB32=C:\Program Files (x86)\Microsoft SDKs\MPI\Lib\x86\
MSMPI_LIB64=C:\Program Files (x86)\Microsoft SDKs\MPI\Lib\x64\

You can check your installation with:

python examples/example_fastfarm.py

On Windows, it is necessary to launch the example directly with mpiexec:

mpiexec -n 1 python examples/example.py

More details on interfacing with FAST.Farm

A simple tutorial to start a simulation with the FastFarm interface is available in the notebook interface.ipynb notebook. To properly launch the notebook, see the intructions below in Running Examples Notebook.

Creating an interface from a WFCRL case:

from wfcrl.environments import data_cases as cases
from wfcrl.interface import FastFarmInterface

config = cases.fastfarm_6t
interface = FastFarmInterface(config)

On Windows, by default, your FAST.Farm executable is assumed to be located in simulators/fastfarm/bin/FAST.Farm_x64_OMP_2023.exe. If not, you can also pass it to the interface:

interface = FastFarmInterface(config, fast_farm_executable=path_to_exe)

Creating an interface from existing configuration files: Alternatively, if you already have your simulation fils ready, you can just point towards the .fstf file:

ff_interface = FastFarmInterface(fstf_file=path_to_fstf)

At every iteration, the FAST.Farm interace retrieves 12 measures per turbine:

• 2 wind measurements: wind velocity and direction at the entrance of the farm
• The current output power of the turbine
• The yaw of the turbine
• The pitch of the turbine
• The torque of the turbine
• 6 measures of blade loads

A detailed example can be found in the interface.ipynb notebook. To run this notebook, follow the instructions under Running Example Notebooks.

Running Example Notebooks

On Windows, to run the interface.ipynb and demo.ipynb examples, you will first need to install the WFCRL kernel:

• Install jupyter notebook and seaborn:

pip install notebook seaborn

• Install the jupyter kernel

from wfcrl import jupyter_utils
jupyter_utils.create_ipykernel()