SEER: Transformer-based Robust Time Series Forecasting via Automated Patch Enhancement and Replacement

Time series forecasting is important in many fields that require accurate predictions for decision-making. Patching techniques, commonly used and effective in time series modeling, help capture temporal dependencies by dividing the data into patches. However, existing patch-based methods fail to dynamically select patches and typically use all patches during the prediction process. In real-world time series, there are often low-quality issues during data collection, such as missing values, distribution shifts, anomalies and white noise, which may cause some patches to contain low-quality information, negatively impacting the prediction results. To address this issue, this study proposes a robust time series forecasting framework called $\textbf{SEER}$. Firstly, we propose an $\textit{Augmented Embedding Module}$, which improves patch-wise representations using a Mixture-of-Experts~(MoE) architecture and obtains series-wise token representations through a channel-adaptive perception mechanism. Secondly, we introduce a $\textit{Learnable Patch Replacement Module}$, which enhances forecasting robustness and model accuracy through a two-stage process: 1) a dynamic filtering mechanism eliminates negative patch-wise tokens; 2) a replaced attention module substitutes the identified low-quality patches with global series-wise token, further refining their representations through a causal attention mechanism. Comprehensive experimental results demonstrate the SOTA performance of SEER.

Time series forecasting is widely used in areas such as energy management, transportation, healthcare, and finance, where accurate predictions are essential for planning and decision-making. Many modern forecasting systems divide historical data into smaller segments, called patches, to better understand temporal patterns. However, existing methods usually treat all patches equally, even though some parts of the data may be unreliable due to missing values, noisy measurements, abnormal events, or changing data distributions. As a result, low-quality information can negatively affect forecasting performance. To address this problem, we propose a robust forecasting framework called SEER. The key idea of SEER is to automatically identify unreliable parts of the data and reduce their negative impact during prediction. Instead of blindly using every patch, SEER learns to filter out low-quality information and replace it with more reliable global representations learned from the overall time series. This allows the model to make more stable and accurate predictions even when the input data is noisy or incomplete. Experiments on multiple real-world datasets demonstrate that SEER consistently outperforms existing forecasting methods and achieves state-of-the-art performance under challenging data conditions.