Abstract: The subsynchronous oscillation caused by the interaction between flexible direct converter station and wind farm threatens the safe and stable operation of power grid. Deep reinforcement learning can effectively cope with the variable operating conditions of wind power flexible direct grid-connected system. An oscillation suppression method based on deep reinforcement learning is proposed.Firstly,based on the mathematical model and mechanism of the flexible direct grid-connected system,the environmental state set,feasible action set and reward function of the system are designed. Then in order to cope with the dimension disasters caused by continuous current and voltage variables in the designed environment state aggregation,the deep deterministic strategy gradient algorithm is used to explore the action decision. Finally,the effectiveness and robustness of the proposed method are verified in the simulation system under multiple operating conditions. The simulation results show that the proposed method can fully adapt to the multi-wind speed operating conditions of flexible direct grid-connected system of offshore wind power,and can effectively suppress the oscillation in a short time.