Optimising Wind Farms with Hybrid AI
RISE researchers have developed a hybrid approach combining physics-based simulation with machine learning to optimize wind farm layouts, achieving nearly 50 times faster performance than traditional implementations.
Designing efficient wind farms is a complex engineering challenge. The placement of each turbine affects airflow and energy output across the entire site, creating wake effects that are difficult to model accurately.
The Hybrid Approach
At RISE, researchers have developed a hybrid approach that combines physics-based simulation with machine learning to balance accuracy and speed. By re-implementing wind farm simulators in JAX, enabling GPU acceleration, vectorisation, and automatic differentiation, the team created a differentiable simulator that runs nearly 50 times faster than traditional implementations.
Benefits
This speed-up enables efficient, gradient-based optimisation of wind farm layouts, resulting in more stable and energy-efficient solutions. Machine learning components are then used to learn complex turbulence effects that are difficult to capture with physics alone.
Broader Applications
Beyond wind energy, the approach can be applied to other engineering domains that rely on mechanistic simulations, demonstrating the power of combining domain knowledge with modern machine learning techniques.
