This paper studies the dimensionality reduction problem for improving time series forecasting using high-dimensional predictors. We propose a novel Supervised Deep Dynamic Principal Component Analysis (SDDP) framework that integrates target variables and lagged observations into the factor extraction process. We construct target-aware predictors by supervising the size of the original predictors using a temporal neural network, assigning greater weight to predictors with strong predictive power. We then perform PCA on the target-aware predictors to extract estimated SDDP factors. This supervised factor extraction not only improves the predictive accuracy of subsequent forecasting tasks but also generates more interpretable and target-specific latent factors. Based on SDDP, we propose a factor-augmented nonlinear dynamic forecasting model that integrates a wide range of factor model-based forecasting approaches. To further demonstrate the broad applicability of SDDP, we extend our research to more challenging scenarios where predictors are only partially observable. We empirically validate the performance of the proposed method on several real-world public datasets. The results demonstrate that the proposed algorithm significantly improves forecasting accuracy compared to state-of-the-art methods.
