深度强化学习驱动下桥梁主动吸气流动控制

An active suction flow control method driven by a deep reinforcement learning (DRL) algorithm was developed to improve the aerodynamic performances of bridges in wind fields. The suction control strategy based on the DRL algorithm can be flexibly adjusted according to the wind field environment. Considering the position of the vortex shedding at the trailing edge of the bridge, the suction slot was set at the corner of the trailing edge of the bottom deck relative to the flow direction. Active suction control in a two-dimensional wind field environment was studied using the coupled DRL and computational fluid dynamics platform DRLinFluids. The results indicate the following. Under suction control based on the DRL algorithm, the fluctuating lift, fluctuating drag, and fluctuating bending moment of the bridge are 99. 2%, 92. 9%, and 98. 5%, respectively, of those without control. Simultaneously, the time-averaged drag and bending moment of the bridge are reduced by 21. 3% and 98. 3%, respectively. (2) Under DRL-based suction control, the vortex shedding in the bridge wake is suppressed, and the wake vorticity changes from a periodic alternating shedding vortex to a stable long tail wake, changing the vortex-shedding location. (3) The suction strategy driven by DRL gradually changes from large-amplitude fluctuations in the early stage to small-amplitude fluctuations. Moreover, the large-amplitude fluctuation time with high energy consumption coincides with the control time for flow field improvement. After the flow field reaches a stable stage, the strategy changes to a small amplitude with a low energy-consumption suction state.