DiffWake: A General Differentiable Wind Farm Solver in JAX

Maria Bånkestad, Leon René Sütfeld, Aleksis Pirinen, Hamidreza Abedi

Keywords: Differentiable programming, physics-informed machine learning, wind-farm flow modeling, cumulative–curl wake model, gradient-based optimization, turbulence intensity calibration, GPU-accelerated simulation, JAX

TL;DR: A differentiable wind-farm flow solver in JAX that preserves wake physics and enables gradient-based optimization and parameter inference.

Abstract:

We introduce DiffWake, an end-to-end differentiable and curl-conserving flow solver for wind farms implemented in JAX.
It preserves the core physics of wake formation and recovery while enabling exact gradient backpropagation through wake, thrust, and power calculations on GPUs. Unlike traditional engineering models with fixed empirical parameters, DiffWake offers a general, differentiable foundation for wind farm simulation that supports optimization, calibration, and machine learning-enhanced modeling. We demonstrate its use for (i) layout optimization under spatial constraints and (ii) probabilistic turbulence calibration from SCADA data using a lightweight neural network. Together, these results demonstrate a unified framework linking physical modeling with machine learning for accurate and scalable wind farm simulation.

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Submission Number: 49