摘要
针对基于点距离的时间序列相似性搜索算法鲁棒性较差的问题,提出一种面向形态的时间序列近似表示方法和相似性度量算法。算法不依赖于时间序列长度和领域知识。在充分利用时间序列时变特征的基础上,以角点为分界点,利用角点处的弯曲度提取时间序列的特征,近似表示时间序列。实验结果表明,该算法具有良好的平移和伸缩不变性及较好的鲁棒性,搜索能力更强。
Aiming at the lack of similar sub-patterns discovery algorithm from time series based on points distance such as poor robustness, an algorithm is proposed for similarity measure and approximate representation of time series based on morphological character. Fully used the time-varying characteristics, time series are divided by angular points, and its characters are extracted with bending degrees at angular points to approximate the time series. The algorithm does not depend on the length of time series and domain knowledge. Experimental results show that this algorithm is not only invariant to translation and scalability, but has good robustness, and the results are more effective.
出处
《计算机工程》
CAS
CSCD
北大核心
2011年第15期37-39,54,共4页
Computer Engineering
基金
国家自然科学基金资助项目(50674086)
关键词
时间序列
相似性搜索
角点弯曲度
形态特征
平滑
time series
similarity search
bending degree of angular point
morphological characteristic
smooth