Scholarlink

<p>Adaptively Constrained Dynamic Time Warping (ACDTW) is a novel algorithm based on Dynamic Time Warping (DTW) designed to address the “singularities” issue commonly found in traditional DTW methods. However, ACDTW requires additional weight computations and uses a matrix to record the usage frequency of each point, significantly increasing computational steps and reducing efficiency. To enhance computational efficiency, this paper proposes a more efficient algorithm, Fast-ACDTW. By optimizing the assignment strategy of the cumulative usage matrix in ACDTW, Fast-ACDTW reduces the number of matrix operations. Additionally, it adopts differentiated computation methods for equal-length and unequal-length sequences to further reduce the computational cost. Furthermore, this paper integrates Fast-ACDTW with lower bound functions and global constraints, proposing the LG-Fast-ACDTW algorithm to enhance efficiency further. Experimental results on the UCR time series dataset show that LG-Fast-ACDTW maintains the same classification accuracy as ACDTW while achieving up to a 91.49% improvement in computational efficiency on specific datasets. The research results demonstrate that the LG-Fast-ACDTW algorithm, by incorporating lower bound functions and global constraints, significantly enhances computational performance and offers strong practical value.</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

LG-Fast-ACDTW: a fast adaptive dynamic time warping algorithm

  • 期刊论文

  • 发表日期：

    2026年6月22日

查看全文
  • Danyang Cao,
  • Hanyu Cui,
  • Di Liu,
  • Zifeng Lin,
  • Lei Yang

摘要

Adaptively Constrained Dynamic Time Warping (ACDTW) is a novel algorithm based on Dynamic Time Warping (DTW) designed to address the “singularities” issue commonly found in traditional DTW methods. However, ACDTW requires additional weight computations and uses a matrix to record the usage frequency of each point, significantly increasing computational steps and reducing efficiency. To enhance computational efficiency, this paper proposes a more efficient algorithm, Fast-ACDTW. By optimizing the assignment strategy of the cumulative usage matrix in ACDTW, Fast-ACDTW reduces the number of matrix operations. Additionally, it adopts differentiated computation methods for equal-length and unequal-length sequences to further reduce the computational cost. Furthermore, this paper integrates Fast-ACDTW with lower bound functions and global constraints, proposing the LG-Fast-ACDTW algorithm to enhance efficiency further. Experimental results on the UCR time series dataset show that LG-Fast-ACDTW maintains the same classification accuracy as ACDTW while achieving up to a 91.49% improvement in computational efficiency on specific datasets. The research results demonstrate that the LG-Fast-ACDTW algorithm, by incorporating lower bound functions and global constraints, significantly enhances computational performance and offers strong practical value.