Scholarlink

<p>Time series analysis plays a critical role in applications such as weather forecasting, anomaly detection, and action recognition. However, modeling non-stationary time series remains a challenging task due to their complex, time-varying patterns. Motivated by these challenges, this paper presents a novel series decomposition framework that extracts periodic components through a learnable period matching mechanism and captures smooth trend variations using ordinary differential equation (ODE)-based neural networks. We introduce the ODEMixer model, which incorporates a fragment similarity calculation module to effectively capture multi-periodic signals. Extensive experiments conducted on the MIT-BIH polysomnographic database demonstrate that ODEMixer achieves superior performance in modeling complex non-stationary time series.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

ODEMixer: an approach for modeling non-stationary time series with learnable fragment library

  • 期刊论文

  • 发表日期：

    2025年10月18日

查看全文
  • Bin Wei,
  • Jiejie Chen,
  • Ping Jiang,
  • Zhiwei Xiao

摘要

Time series analysis plays a critical role in applications such as weather forecasting, anomaly detection, and action recognition. However, modeling non-stationary time series remains a challenging task due to their complex, time-varying patterns. Motivated by these challenges, this paper presents a novel series decomposition framework that extracts periodic components through a learnable period matching mechanism and captures smooth trend variations using ordinary differential equation (ODE)-based neural networks. We introduce the ODEMixer model, which incorporates a fragment similarity calculation module to effectively capture multi-periodic signals. Extensive experiments conducted on the MIT-BIH polysomnographic database demonstrate that ODEMixer achieves superior performance in modeling complex non-stationary time series.