In recent years, sub-synchronous oscillation (SSO) phenomena occur in direct-drive permanent magnet synchronous generator (D-PMSG) based wind power generation systems (WPGS) in many regions of China. The SSO characteristics, frequency and damping, are all time-varying mostly affected by changes of wind speed and D-PMSGs number. It is the importance for the design of SSO suppression to know the variation ranges of characteristics and the risk of unstable SSO. The small signal model of the D-PMSG-based WPGS is firstly established in this paper, considering the functional relationship between wind speed and wind turbine output power. Based on the small signal model and the probability distribution function (PDF) of wind speed, the SSO probability assessment method with the least square method (LSM) polynomial fitting is then proposed. Therefore, the probability distributions of SSO frequency and damping considering wind speed variation are obtained efficiently. By PSCAD/EMTDC simulations and comparisons, the obtained results with the eigenvalue analysis method based on the small signal model are certificated. Then the probability assessment method is applied to study the influences of different average wind speeds under different number of D-PMSGs on the probability distribution of SSO frequency and damping. The changes of SSO frequency bandwidth and the risk of unstable SSO are depicted relative completely. The variation range of SSO frequency is consistent with the actual accident scenario in Hami area of Xinjiang, China.INDEX TERMS Direct-drive permanent magnet synchronous generator (D-PMSG), probability assessment, sub-synchronous oscillation (SSO), wind speed.