Abstract: In order to solve the problem of flow separation phenomenon caused by high-angle-of-attack airfoils under weak turbulent conditions, a soft actor-critic (SAC) deep reinforcement learning (DRL) algorithm was used to train a neural network for closed-loop active flow control strategy. In complex environments, an optimized strategy was developed by introducing a zero-mass jet constraint and utilizing three jets, obtaining average drag coefficient reductions for different angles of attack. Results show that using the DRL training strategy to control the jet velocity can inhibit the flow separation of airfoil effectively. When single-jet control is used and airfoil attack angles are 10°, 13° and 15°, the average drag coefficients are reduced by 25%, 15.3% and 11.7%, respectively. Under the condition of large angle of attack, the DRL-based active flow control method has a good effect, which verifies high efficiency of the method in restraining flow separation of airfoil. The introduction of the synthetic jet also enables the agent to find a better control strategy and further reduce the drag coefficient.