---
id: 20260304-T0-14
title: "StaTS模型实现自适应时间序列预测"
title_en: "StaTS Model Enables Adaptive Time Series Prediction"
url: https://ai.daily.yangsir.net/daily/20260304-T0-14
issue_date: 2026-03-04
publish_date: 2026-03-03T05:00:00.000Z
source_name: "arXiv cs.LG (ML)"
source_url: https://arxiv.org/abs/2603.00037
---

# StaTS模型实现自适应时间序列预测

研究者提出StaTS方法，结合频域引导降噪器提升时间序列预测精度。该模型通过谱轨迹调度学习，能根据数据动态调整噪声衰减策略，在气象预测和电力负荷预测任务中，均方误差降低18%。相比传统扩散模型，StaTS的中间状态可逆性提升70%，预测结果更接近真实分布。代码已开源，支持PyTorch框架。

## English Version

**StaTS Model Enables Adaptive Time Series Prediction**

Researchers propose the StaTS method, combining a frequency-domain guided denoiser to improve time series prediction accuracy. The model learns through spectral trajectory scheduling to dynamically adjust noise decay strategies based on data, reducing mean squared error by 18% in weather forecasting and power load prediction tasks. Compared to traditional diffusion models, StaTS improves intermediate state reversibility by 70%, making predictions closer to the true distribution. Code is open-sourced and supports PyTorch.

---

**来源**：[arXiv cs.LG (ML)](https://arxiv.org/abs/2603.00037)

**详情页**：https://ai.daily.yangsir.net/daily/20260304-T0-14

---

*智语观潮 · Daily — https://ai.daily.yangsir.net/llms.txt*