Extreme Adaptive Transformer for Time Series Forecasting
中文标题: 极端自适应Transformer用于时间序列预测
英文摘要
The Extreme-Adaptive Transformer (Exformer) is proposed for hydrologic streamflow forecasting, targeting the underrepresentation of rare extreme events in traditional Transformers. Its attention mechanism consists of three sparse components: Local (short-term), Stride (periodic), and Extreme (event-aware dependencies between normal and extreme patterns). Evaluated on four real-world hydrologic streamflow datasets, Exformer outperforms state-of-the-art baselines on 3-day forecasting. The results show that explicitly incorporating extreme-aware attention improves Transformer models on imbalanced time series with critical rare events.
中文摘要
提出极端自适应Transformer（Exformer）用于水文流量预测，解决传统Transformer对罕见极端事件表征不足的问题。其注意力机制包含三个稀疏成分：局部（短期）、步幅（周期）和极端（普通与极端模式间的事件感知依赖）。在四个真实水文流量数据集上，Exformer在3日预测任务上超越了现有最优基线。结果表明，明确加入极端感知注意力能提升Transformer在不平衡时间序列（含关键罕见事件）上的预测能力。
关键要点
Exformer, an Extreme-Adaptive Transformer, explicitly models both normal and extreme temporal dependencies for time series forecasting.
Exformer（极端自适应Transformer）显式建模正常与极端时序依赖，用于时间序列预测。
The model introduces an extreme-adaptive attention with Local, Stride, and Extreme sparse components, where the Extreme component captures event-aware dependencies between normal and extreme streamflow patterns.
模型引入极端自适应注意力，包含局部、步幅和极端三个稀疏成分，其中极端成分捕获正常与极端流量模式间的事件感知依赖。
Experiments on four real-world hydrologic streamflow datasets show Exformer achieves superior 3-day forecasting performance over state-of-the-art baselines.
在四个真实水文流量数据集上的实验表明，Exformer在3日预测任务上性能优于现有最优基线。
The approach improves Transformer forecasting on imbalanced time series where rare but consequential extreme events are critical.
该方法提升了Transformer在极端事件稀少但影响巨大的不平衡时间序列上的预测能力。