基于深度强化学习的翼型流动分离自适应主动控制研究
Adaptive active control research of airfoil flow separation based on deep reinforcement learning
摘要
为削弱翼型大迎角状态下因边界层流动分离导致的性能消极影响,针对Re=1.6×105下NACA0012翼型,以等离子体激励为流动控制手段,利用深度强化学习算法,依据流场反馈的流动信息,对翼型表面的激励射流进行实时自适应闭环控制,实现静止、动态俯仰翼型后缘流动分离控制.计算结果表明:离散型动作和连续性动作的智能体均能在与流场的若干次交互中探索到控制流动分离的有效策略;奖励函数的动作惩罚项参数大小影响智能体在追求目标和避免惩罚之间的取舍程度;不同目标值的奖励函数能够引导智能体实现对翼型后缘分离状态的精准控制,在一定范围内控制误差在6%以下.对动态俯仰翼型后缘流动分离控制,无惩罚智能体通过在较大迎角阶段的持续激励,延迟翼型后缘负压区发展,消除了后续的负压区,可带来高达92%的升力增益;有惩罚智能体在负压区向下游移动期间施加激励,有利于削弱负压区的负压峰值及提前流动再附着.在泛化测试中,智能体对于缩减频率增大的翼型控制效果更好.
To mitigate the negative impact on aerodynamic performance caused by boundary layer flow separation at high angles of attack in airfoil,the article employs plasma excitation as a flow control method for the NACA0012 airfoil at Re=1.6×105.Utilizing a deep reinforcement learning algorithm,real-time adaptive closed-loop control of excitation jets on the airfoil surface was achieved based on feedback flow information from the flow field.This was applied to control flow separation at the trailing edge of static and pitching airfoil.The computational results indicate that both discrete and continuous action agents can explore effective strategies for controlling flow separation through several interactions with the flow field.The size of the action penalty parameter in the reward function affects the agent's trade-off between pursuing the objective and avoiding penalties.Reward functions with different target values can guide agents to achieve precise control of the trailing edge separation state,maintaining control errors below 6%within a certain range.For controlling trailing edge flow separation of pitching airfoil,the penalty-free agent achieves up to 92%lift gain by continuously exciting during high angle of attack stages,delaying the development of the trailing edge negative pressure zone and eliminating subsequent negative pressure zones.The penalized agent applies excitation during the downstream movement of the negative pressure zone,conducing to weaken the negative pressure peak and promote early flow reattachment.In generalization tests,agents exhibit better control effects on airfoil with increased pitching frequencies.
年,卷(期):
2026, 45(20)
分类号:
O355
机标分类号:
V211.3
TB535
TP273+.2
在线出版日期:
2026年11月3日
基金项目:
12372232:国家自然科学基金,