Real-time actuator line simulations of wind farm flows enabled by the lattice Boltzmann method and GPUs

This study investigates the feasibility of achieving real-time performance in high-fidelity wind farm simulations using the actuator line model (ALM). While such large eddy simulations have traditionally been considered too computationally demanding, we address this challenge by utilizing the lattice Boltzmann method (LBM) accelerated by multi-graphics processing unit (multi-GPU) computing. By employing a computation–communication overlap technique, the proposed approach achieves excellent strong scaling performance, maintaining a parallel efficiency of 97.4% from 2 to 8 GPUs for a system comprising 1.42 billion grid points while attaining a high execution throughput of 8,308 million lattice updates per second per GPU. Real-time performance evaluations for a wind farm consisting of twelve 15 MW turbines confirm that real-time capability is achieved at standard ALM resolutions of 32 and 48 points per rotor diameter, with execution-to-physical time ratios of 0.38 and 0.79, respectively. Even at a higher resolution of 64 points per diameter, the ratio remains 1.41, indicating that high-resolution ALM analyses are likely to be achievable in real time on next-generation GPUs. Furthermore, weak scaling evaluations demonstrate a parallel efficiency of 86% from 4 to 32 GPUs, confirming that real-time performance can be sustained for larger wind farms by increasing the number of GPUs in proportion to the number of turbines. These results indicate that LBM-based simulations constitute a powerful tool for real-time wind farm flow prediction.