The classification of disturbance event types and the identification of their regional locations are critical for the stability analysis and control of interconnected power grids. To address this, a disturbance event classification and localization method based on the S-Koopman energy matrix is proposed. This method uses the subspace optimal mode decomposition algorithm (Sub-OpMD) to calculate the optimal low dimensional matrix of the S-Koopman operator, which captures the dynamic information of the stochastic system. The S-Koopman energy matrix is then constructed according to the eigen decomposition results of the optimal low dimensional matrix. The S-Koopman energy matrix reflects the dynamic behavior of the system; its row/column vectors can extract different feature data following disturbance events, as well as the fluctuation strengths of PMU data at different positions. Furthermore, by analyzing the dynamic variation patterns of PMU data under different disturbance types and regions, quantitative indicators of disturbance event classification and regional localization are proposed. The calculation and analysis of the New England 10-generator 39-bus simulation system and the CEPRI-SSFS 197-bus real-world system verify the rationality, effectiveness, and accuracy of the proposed method.