The Double-Window Principal Component Analysis (DWPCA) algorithm introduces both temporal and spatial windows on the basis of traditional PCA algorithm, using a moving temporal window to effectively distinguish different structural states and employing a spatial window to eliminate data insensitive to damage. However, in the DWPCA algorithm, the size of the temporal window is fixed, typically chosen as 1-2 times the maximum period of response. The determination of the temporal window scale and its impact on damage recognition capability for different types of damage remains to be explored. This paper firstly introduces temporal windows of different scales (yearly/seasonal/monthly) into DWPCA to investigate the evolution patterns of PCA features over time under different temporal windows. Secondly, PCA features corresponding to different temporal windows are utilized as inputs for Machine Learning (ML) algorithms, combined with ML algorithms for structural damage localization and quantification tasks. Finally, the effectiveness of the method is validated using a simulation model of a double-span continuous beam and a Benchmark model. Results indicate that yearly temporal windows exhibit good robustness to damage, while monthly temporal windows demonstrate higher sensitivity to damage. The synergistic effect of the three types of temporal windows leverages their respective advantages, further enhancing the damage localization and quantification capabilities of ML.