Abstract: The static var generator(SVG), with its rapid dynamic response characteristics, plays a significant role in suppressing sub-synchronous oscillations during the grid integration of doubly-fed wind power systems. However, traditional control strategies still have certain limitations when dealing with the system’s complex nonlinear and time-varying behaviors. To address this, a sub-synchronous oscillation suppression method is proposed based on empirical mode decomposition(EMD), kernel principal component analysis(KPCA), long short-term memory(LSTM), and the additional damping control of SVG. First, the oscillatory features of the system are extracted through EMD, followed by dimensionality reduction optimization using KPCA. Then, the dynamic characteristics of the system are modeled and predicted using LSTM, significantly improving prediction accuracy. On this basis, the additional damping control function of SVG is utilized to adjust the control signals in real time, effectively suppressing sub-synchronous oscillations and enhancing system stability. The innovation of this proposed method lies in combining signal processing techniques with deep learning algorithms, constructing an efficient prediction and control framework, and providing a novel approach to traditional control strategies. Finally, simulation analysis using PSCAD validates the effectiveness of the proposed method, offering technical support for the stable operation of high-penetration renewable energy grids.