Deep Reinforcement Learning (DRL) excels in active flow control, yet the existing studies often rely on simplified flow configurations and low Reynolds number conditions deviating from real applications. This work applies Proximal Policy Optimization (PPO) to control distributed Dual Synthetic Jets (DSJs) for lift enhancement on a NACA0015 airfoil at high angle of attack and Re = 250000. A near-practical configuration integrates non-uniform surface pressure and near-wall velocity measurements, while the actuation process of DSJs is accurately resolved with a numerical simulation time step of 5×10 -5 s. The trained PPO agent dynamically optimizes DSJ velocity distribution under energy constraints, establishing a relay-control mechanism that suppresses flow separation and converges to optimal actuation. This achieves 80.8% drag reduction and 100.4% lift increase (20.7% improvement over open-loop control). The strategy maintains robust performance across varying Reynolds numbers and time-varying angles of attack, demonstrating strong dynamic control capabilities. This study successfully implements DRL for distributed DSJ-based separation control in turbulent environments of high Reynolds number with near-experimental configurations, providing critical insights for experimental and engineering applications.
Liu et al. (Sun,) studied this question.