用无标签数据训练时间序列事件预测模型，少标注也能准
HEPA: A Self-Supervised Horizon-Conditioned Event Predictive Architecture for Time Series
- 用未来表示预测代替直接预测值，自监督学习捕捉时序规律
- 在14个基准上10项超越主流模型，标签数据少10倍仍有效
- 适合标签稀缺的工业、医疗等高风险事件预警场景
多变量时间序列中的关键事件（如涡轮机故障、心律失常）需精准预测，但标注数据稀缺，因事件罕见且标注成本高。我们提出HEPA（Horizon-conditioned Event Predictive Architecture），基于两大原则：首先，使用因果Transformer编码器，通过联合嵌入预测架构（JEPA）进行自监督预训练——以未来表示而非具体数值为目标进行预测，迫使编码器仅从无标签数据中学习可预测的时序动态；其次，冻结编码器，仅微调预测器以生成目标事件的单调生存累积分布函数（CDF）。在全部11个领域共14个基准测试中，使用固定架构与优化器超参数，HEPA在水污染、网络攻击检测、波动率状态识别等8种事件类型上表现优异，超越PatchTST、iTransformer、MAE和Chronos-2等主流模型至少10项，参数量减少一个数量级，生命周期数据集上所需标注数据也少一个数量级。
原文摘要 · Abstract (English)
Critical events in multivariate time series, from turbine failures to cardiac arrhythmias, demand accurate prediction, yet labeled data is scarce because such events are rare and costly to annotate. We introduce HEPA (Horizon-conditioned Event Predictive Architecture), built on two key principles. First, a causal Transformer encoder is pretrained via a Joint-Embedding Predictive Architecture (JEPA): a horizon-conditioned predictor learns to forecast future representations rather than future values, forcing the encoder to capture predictable temporal dynamics from unlabeled data alone. Second, we freeze the encoder and finetune only the predictor toward the target event, producing a monotonic survival cumulative distribution function (CDF) over horizons. With fixed architecture and optimiser hyperparameters across all benchmarks, HEPA handles water contamination, cyberattack detection, volatility regimes, and eight further event types across 11 domains, exceeding leading time-series architectures including PatchTST, iTransformer, MAE, and Chronos-2 on at least 10 of 14 benchmarks, with an order of magnitude fewer tuned parameters and, on lifecycle datasets, an order of magnitude less labeled data.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。