L-Drive: Beyond a Single Mapping—Latent Context Drives Time Series Forecasting

Mainstream methods for multivariate time-series forecasting largely follow the Direct-Mapping paradigm. They learn a unified mapping from history to the future in the observation space to fit value-level dependencies. However, real-world systems often undergo distribution shifts and regime changes. In such cases, a unified mapping can exhibit response lag around turning points, causing error accumulation within the switching window and reducing forecasting reliability. To address this issue, we propose L-Drive, a change-aware forecasting framework. L-Drive introduces a Latent-Context, to explicitly characterize high-level dynamics evolving over time, and uses gating to modulate increment representations. This provides more timely change cues and improves adaptation to changing segments. In addition, it incorporates patch-shared relative positional basis functions to strengthen intra-segment structural modeling and reduce overfitting caused by absolute-position memorization. Extensive experiments validate the effectiveness of L-Drive and show a better overall trade-off between forecasting accuracy and computational efficiency.

Predicting the future behavior of complex systems such as economic indicators or climate patterns is highly challenging because their dynamics often evolve over time. Traditional forecasting methods typically rely on a fixed mapping from past to future values, which can be slow to respond or less accurate when the data experiences sudden changes or new trends. To address this problem, we propose L-Drive, a forecasting framework that incorporates a Latent-Context (L-Context). L-Context is an internal representation that captures the continuously evolving dynamics of the system. This latent context enables the model to quickly detect changes in behavior and make more accurate predictions. In addition, L-Drive leverages shared relative positional structures within data segments, helping the model learn local patterns while avoiding overfitting to exact positions. Experiments show that L-Drive outperforms traditional methods, achieving a good balance between prediction accuracy and computational efficiency. It enhances the adaptability and predictive performance of time series in dynamic environments and provides new ideas for more complex real-world applications.