Reinforcement Learning–Based Wind Farm Layout Optimization Using Neural Surrogate Models and Real Wind Data

Ali Zaidi
Proceedings of the The 39th Canadian Conference on Artificial Intelligence, PMLR 318:747-758, 2026.

Abstract

This paper presents a data-driven framework for wind farm layout optimization that integrates neural surrogate modeling and reinforcement learning to address complex wake interactions and high computational costs associated with physics-based simulations. A deep neural network surrogate is trained on Reynolds-Averaged Navier–Stokes (RANS) simulation data to predict turbine power output from localized flow velocity features sampled within a 1.5 rotor-diameter neighborhood. Validation on a held-out test dataset shows normalized prediction errors below 5% when benchmarked against both manufacturer-derived power curves and RANS CFD results. The surrogate is embedded within a reinforcement learning framework to perform sequential, wake-aware turbine placement using realistic wind data for Winnipeg, Canada. The proposed approach achieves stable convergence and produces denser, higher-performing layouts than a genetic algorithm baseline under identical constraints, demonstrating its effectiveness for scalable wind farm optimization.

Cite this Paper

BibTeX
@InProceedings{pmlr-v318-zaidi26a, title = {Reinforcement Learning–Based Wind Farm Layout Optimization Using Neural Surrogate Models and Real Wind Data}, author = {Zaidi, Ali}, booktitle = {Proceedings of the The 39th Canadian Conference on Artificial Intelligence}, pages = {747--758}, year = {2026}, editor = {Bouzar-Benlabiod, Lydia and Leung, Carson}, volume = {318}, series = {Proceedings of Machine Learning Research}, month = {25--29 May}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v318/main/assets/zaidi26a/zaidi26a.pdf}, url = {https://proceedings.mlr.press/v318/zaidi26a.html}, abstract = {This paper presents a data-driven framework for wind farm layout optimization that integrates neural surrogate modeling and reinforcement learning to address complex wake interactions and high computational costs associated with physics-based simulations. A deep neural network surrogate is trained on Reynolds-Averaged Navier–Stokes (RANS) simulation data to predict turbine power output from localized flow velocity features sampled within a 1.5 rotor-diameter neighborhood. Validation on a held-out test dataset shows normalized prediction errors below 5% when benchmarked against both manufacturer-derived power curves and RANS CFD results. The surrogate is embedded within a reinforcement learning framework to perform sequential, wake-aware turbine placement using realistic wind data for Winnipeg, Canada. The proposed approach achieves stable convergence and produces denser, higher-performing layouts than a genetic algorithm baseline under identical constraints, demonstrating its effectiveness for scalable wind farm optimization.} }
Endnote
%0 Conference Paper %T Reinforcement Learning–Based Wind Farm Layout Optimization Using Neural Surrogate Models and Real Wind Data %A Ali Zaidi %B Proceedings of the The 39th Canadian Conference on Artificial Intelligence %C Proceedings of Machine Learning Research %D 2026 %E Lydia Bouzar-Benlabiod %E Carson Leung %F pmlr-v318-zaidi26a %I PMLR %P 747--758 %U https://proceedings.mlr.press/v318/zaidi26a.html %V 318 %X This paper presents a data-driven framework for wind farm layout optimization that integrates neural surrogate modeling and reinforcement learning to address complex wake interactions and high computational costs associated with physics-based simulations. A deep neural network surrogate is trained on Reynolds-Averaged Navier–Stokes (RANS) simulation data to predict turbine power output from localized flow velocity features sampled within a 1.5 rotor-diameter neighborhood. Validation on a held-out test dataset shows normalized prediction errors below 5% when benchmarked against both manufacturer-derived power curves and RANS CFD results. The surrogate is embedded within a reinforcement learning framework to perform sequential, wake-aware turbine placement using realistic wind data for Winnipeg, Canada. The proposed approach achieves stable convergence and produces denser, higher-performing layouts than a genetic algorithm baseline under identical constraints, demonstrating its effectiveness for scalable wind farm optimization.
APA
Zaidi, A.. (2026). Reinforcement Learning–Based Wind Farm Layout Optimization Using Neural Surrogate Models and Real Wind Data. Proceedings of the The 39th Canadian Conference on Artificial Intelligence, in Proceedings of Machine Learning Research 318:747-758 Available from https://proceedings.mlr.press/v318/zaidi26a.html.

Related Material