Scholarlink

<p>With the rapid development of the Internet of Things (IoT), massive numbers of devices generate vast amounts of time series data. Efficient compression techniques are essential for reducing storage costs and enhancing query performance. Among various data types, floating-point time series data poses significant compression challenges due to its high precision and dynamic range. However, the numerical patterns of time series data vary widely across applications and sensor types, so compression algorithms designed for a single pattern often lack the flexibility to address diverse needs. To overcome this limitation, we propose NPAC, a <i>N</i>umeric <i>P</i>attern-<i>A</i>ware <i>C</i>ompression algorithm for floating-point time-series data. NPAC employs a classification model at the time window level to identify the underlying numerical patterns in the data and introduces a two-layer decision architecture to balance compression ratio and time overhead. In the first layer, several candidate schemes with potentially high compression ratios are selected based on the classification model. In the second layer, a cost function evaluates both compression and decompression time to select the most efficient scheme as the optimal one. We evaluate NPAC against traditional compression algorithms using 18 real-world datasets. The results show that NPAC outperforms the others in most scenarios, achieving an average compression ratio improvement of 61.63%.</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

NPAC: numeric pattern aware compression algorithm for floating-point time-series data

  • 期刊论文

  • 发表日期：

    2025年10月27日

查看全文
  • Wenjing Wang,
  • Liang Liu,
  • Kaibin Zhang,
  • Keyue Yang,
  • Lingwei Kuang,
  • Ziyi Zheng,
  • Jinan Wang,
  • Junliang Cao

摘要

With the rapid development of the Internet of Things (IoT), massive numbers of devices generate vast amounts of time series data. Efficient compression techniques are essential for reducing storage costs and enhancing query performance. Among various data types, floating-point time series data poses significant compression challenges due to its high precision and dynamic range. However, the numerical patterns of time series data vary widely across applications and sensor types, so compression algorithms designed for a single pattern often lack the flexibility to address diverse needs. To overcome this limitation, we propose NPAC, a Numeric Pattern-Aware Compression algorithm for floating-point time-series data. NPAC employs a classification model at the time window level to identify the underlying numerical patterns in the data and introduces a two-layer decision architecture to balance compression ratio and time overhead. In the first layer, several candidate schemes with potentially high compression ratios are selected based on the classification model. In the second layer, a cost function evaluates both compression and decompression time to select the most efficient scheme as the optimal one. We evaluate NPAC against traditional compression algorithms using 18 real-world datasets. The results show that NPAC outperforms the others in most scenarios, achieving an average compression ratio improvement of 61.63%.