Deep reinforcement learning-based active flow control for a tall building

This study focuses on developing a deep reinforcement learning (DRL) flow control policy to mitigate aerodynamic loads of a tall building at high Reynolds number Re=7.53×104. Multiple jets are placed at the four corners and the free end of the building, aiming to suppress wind loading of the building under varying incoming wind conditions. Pressure probes on the building surface are used as feedback observers. The soft actor-critic (SAC) algorithm is deployed to train an effective DRL control policy. The DRL agent can optimize the jet velocities, resulting in reductions of 39.1%, 53.7%, and 38.4% in the fluctuations of the drag, lift, and moment coefficients, respectively. Furthermore, the mean drag coefficient is reduced by 27.3%. This study investigates the behavior of multiple jets and their effects on the wind force and flow field. It was found that the multiple jets can reduce crosswind force fluctuations in tall buildings, enhance downwash flow, and mitigate the shedding of wake vortices. These results highlight the potential of DRL in active flow control and lay the foundation for the efficient, robust, and practical implementation of this control technique in real-world engineering applications.