The invention discloses an improved particle swarm algorithm-based ultra-low-frequency oscillation suppression method. The method comprises the steps of S1, applying same power disturbance on a directcurrent system separately, and recording each ultra-low-frequency oscillation modal amplitude value under disturbance; S2, selecting FC control sensitivity points of the ultra-low-frequency oscillation modal amplitude values; S4, establishing a particle swarm algorithm optimization model of the system; S5, setting each item of initial data of the particle swarm, and setting the current number ofiteration into 1; S6, updating particle speed and positions by adopting the particle swarm algorithm; S7, determining the current best individual extreme value; S8, determining the current best global extreme value; S9, performing nonlinear reducing of inertial weight, and updating the inertial weight value; and S10, if the current number of iteration is equal to the largest number of iteration,outputting an FC parameter optimization result, or otherwise, adding one to the current number of iteration and returning to the step S6. By virtue of the way, ultra-low-frequency oscillation can be suppressed effectively, simply and directly.