Active Control of Flow Past a Near-Wall Cylinder Based on Deep Reinforcement Learning

Based on the deep reinforcement learning (DRL) method a pair of jets with zero mass flow applied to the cylindrical surface was used to study the active flow control of a near-wall cylinder for Re=200, 400, and the gap ratio G/D=0.5, 0.7, 1.0, 1.5, 2.0. The jet control strategies and corresponding control effects under different parameters were obtained by the DRL method, and the control effects of different jet positions (90°, 270°)、(90°, 320°)、(90°, 360°) were discussed. It is found that the Reynolds number, gap ratio and jet position all have important influence on the control effects. When the jet position is (90°, 270°), the drag coefficient and its fluctuation can be effectively reduced by DRL control, the controlled cylinder wake is elongated and the pressure difference between the front and rear of the cylinder is reduced. The control effects of the jet positions (90°, 320°) and (90°, 360°) are similar, which can reduce the average drag coefficient, but the drag coefficient after control fluctuates greatly due to the asymmetry of the control positions. The increase of Reynolds number and gap ratio will increase the mass flow level of the control jet. Under the same conditions, using the jet position of (90°, 270°) can get better control effects with the same mass flow rate.