Abstract: With the continuous development of new power systems in China, the problem of sub-synchronous oscillation in power systems has become prominent, seriously affecting the safe and stable operation of the power grid, and the level of oscillation damping has an important impact on the sub-synchronous oscillation of wind farms. As the system damping changes with the operation mode of the power system, a sub-synchronous oscillation suppression strategy for wind farms based on the deep Q network optimization operation mode was proposed. Firstly, the influence of pitch angle and series compensation capacitor on sub-synchronous oscillation damping of wind farms was analyzed by time domain simulation, and on this basis, a joint optimization mathematical model of sub-synchronous oscillation with adjusting doubly fed induction generator (abbr. DFIG) output by pitch angle and adjusting line series compensation by parallel capacitor was established. Secondly, the deep Q-learning network algorithm was applied to the optimization solution of system oscillation damping to obtain the optimization strategy of wind turbine sub-synchronous oscillation suppression, and the results are compared with the results of sub-synchronous oscillation suppression based on the genetic algorithm, The results show that this method effectively reduces the oscillation amplitude and improves the damping of the system, which verifies the rationality and superiority of this method.