How bdvanberkel/wakewise works
Project type and purpose
• Project type: A Python‐based simulation environment implementing the Gymnasium API for reinforcement learning in wind‐farm control.
• Purpose: To simulate wake steering in wind farms using a Graph Neural Network surrogate, model fatigue degradation & maintenance costs, and sample realistic wind and electricity price conditions.
File structure
• Top‐level entry points:
– run.py (CLI or script to launch experiments)
– WakeWISE/main.py (defines the Gym environment class and simulation loop)
• Modules & directories:
– PropagationModels: GNN surrogate + farm simulation logic
– WindSamplers: Markov‐chain wind direction/speed sampler
– DegradationModels & MaintenanceModels: fatigue damage & cost models
– utils: config loader, logging, data handling, graph conversion, normalization
– wrappers: Gym wrappers that modify observation/action spaces or rewards
• Data/config folders:
– configs/*.yaml – environment and degradation parameters
– data/* – wind‐farm layouts, spot‐price matrices, pretrained models
• Dependencies: pyproject.toml, Gymnasium API, PyTorch (GNN models)
README for additional insights
• Core loop: at each 10‐minute timestep, sample wind & price, run GNN surrogate to get power & DELs, update fatigue, compute maintenance cost, and produce reward for RL agent.
• Wrappers exist to tweak observability, normalization, baseline subtraction, graph encoding, etc.
Final checklist
– Include all major components listed above
– Annotate where external data (layouts, price matrices, model weights) is loaded
– Show the wrapper stack around the base environment