大工至善|大学至真分享 http://blog.sciencenet.cn/u/lcj2212916

博文

[转载]【计算机科学】【2020】基于创新机器学习方法的周期时间序列数据分析

已有 319 次阅读 2021-8-13 18:24 |系统分类:科研笔记|文章来源:转载

图片


本文为加拿大渥太华大学(作者:Haolong Zhang)的硕士论文,共100页。

 

周期长度的提取在许多研究领域都是一个难题。为了解决这个问题,在许多应用中提出了不同的方法。例如,供应链管理是一个可以从精确的周期性信息中获益匪浅的领域。另外,生理数据的周期性信息可以提供对个体健康状况的洞察,这也是本文的动机所在。周期长度提取的难点在于工作环境中噪声水平的变化。在一种环境中表现良好的系统在另一种环境中可能不准确。

 

在这项工作中,我们探讨了两种机器学习方法,两种方法都试图在不同的噪声水平下解决这个问题。第一种算法是周期分类算法(PCA),利用历史标记数据作为训练样本,对新实例进行分类。PCA算法对人工噪声和自然噪声都具有鲁棒性。然而,如果数据中没有太多的噪声,PCA的训练是不经济的。第二种算法是周期检测算法(PDA),是在噪声水平不是很高的情况下使用的。它不需要历史数据,而是直接从数据流中检测周期长度。PDA不能像PCA那样耐受噪声;但是,它更高效,部署更简单。通过对这两种算法在人工数据集和真实数据集上的研究,我们确定了它们在不同情况下的优势。特别是在系统无噪声的情况下,PDA的性能优于PCA,而在通常含有大量噪声的真实数据集上,PDA的性能较差。相比之下,鉴于训练数据只是测试数据集的代表,PCA在人工和真实数据集上都表现出了很高的性能。

 

Period length extraction is considered achallenge in many research fields. To solve this problem, different methodshave been proposed across many applications. For instance, supply chainmanagement is an area that can greatly benefit from precise periodicinformation. In addition, periodic information on physiological data canprovide insights into individuals’ health conditions, which is the motivationof this thesis. The difficulty of period length extraction involves the varyingnoise levels among working environments. A system that performs well in oneenvironment may not be accurate in another. In this work, we explore twomachine learning approaches, each of which attempts to solve the problem at adifferent noise level. The first algorithm, the period classification algorithm(PCA), utilizes historical labeled data as training material and classifies newinstances. The PCA demonstrates robustness to both generated and natural noise.However, the training of the PCA is not economical if the data do not containmuch noise. The second algorithm, the period detection algorithm (PDA), is usedwhen the noise level is not very high. It does not require historical data, butrather detects the period length directly from the data stream. The PDA cannottolerate as much noise as the PCA; however, it is more efficient and simpler todeploy. By investigating both algorithms on artificial and real-world datasets,we determined that they have advantages under different circumstances. Inparticular, the PDA outperforms the PCA when the system is noise-free, while itfails on real-world datasets, which usually contain a large amount of noise. Incontrast, given that the training material is representative of test datasets,the PCA demonstrates high performance on both artificial and real-worlddatasets.

 

1. 引言

2.相关工作

3.基于监督深度卷积神经网络的周期时间序列数据分类

4.基于学习自动机的受控环境下周期长度检测

5.结论与展望


更多精彩文章请关注公众号:205328s611i1aqxbbgxv19.jpg



http://wap.sciencenet.cn/blog-69686-1299627.html

上一篇:[转载]【武器系统】【1994.05】全视GPS用于弹道导弹导航性能和完备性监测
下一篇:[转载]【雷达与对抗】【2013.08】相控阵系统的设计与应用

0

该博文允许注册用户评论 请点击登录 评论 (0 个评论)

数据加载中...

Archiver|手机版|科学网 ( 京ICP备07017567号-12 )

GMT+8, 2021-12-1 00:21

Powered by ScienceNet.cn

Copyright © 2007- 中国科学报社

返回顶部