Due to the increasing penetration of wind generations, a new type of sub-synchronous interactions, the so called sub-synchronous control interaction (SSCI), has emerged in power systems in the past decade. Incidents of this type have caused severe consequences in real-life power systems, including equipment damage and loss of power supply. To overcome the impacts and minimize the adverse influences of such phenomenon, this paper proposes a novel varying-frequency adaptive filter (VFAF) in doubly fed induction generators-based wind farms for SSCI mitigation. The proposed VFAF utilizes a linear prediction model and coefficient-based polynomials to analyse the spectral components of the controller signals. In addition, singular value decomposition is used to identify the number of signal components in the controllers. Then, based on the analysis, the VFAF produces compensation signals to modulate the distorted controller signals and thus mitigates the oscillations during SSCI. Based on this mechanism, the VFAF is auto-adaptive to various oscillation frequencies, including those that are time-varying. The above feature is of particularly importance in modern power systems. Testing cases of the VFAF have been conducted in the Power Factory software, and the results verify the effectiveness of the proposed VFAF under relevant system conditions.