计算机与现代化

• 人工智能 •    下一篇

一种面向工作负载预测的基于小波变换的特征提取方法

  

  1. (1.贵州大学计算机科学与技术学院,贵州贵阳550025;2.贵州省先进计算与医疗信息服务工程实验室,
    贵州贵阳550025;3.中国科学院国家天文台,北京100101)
  • 收稿日期:2020-02-06 出版日期:2020-05-20 发布日期:2020-05-21
  • 作者简介:王可(1991-),男,山东菏泽人,硕士研究生,研究方向:大规模数据分析处理,E-mail: wangke9926@gmail.com; 通信作者:李晖(1982-),男,湖南衡阳人,教授,博士,研究方向:大规模数据管理与分析,数据挖掘,E-mail: cse.huili@gzu.edu.cn; 陈梅(1964-),女,贵州贵阳人,教授,研究方向:数据库技术,数据挖掘,E-mail: gychm@gzu.edu.cn; 戴震宇(1985-),男,贵州贵阳人,实验师,研究方向:大规模数据分析处理,E-mail: zydai@gzu.edu.cn; 朱明(1966-),男,广西玉林人,研究员,博士,研究方向:科学数据处理,射电天文学,E-mail: mz@nao.cas.cn。
  • 基金资助:
    国家自然科学基金资助项目(61462012, 61562010, 71964009); 贵州省高层次创新型人才项目(2017)

A Wavelet Transform-based Feature Extraction Method for Workload Prediction

  1. (1. College of Computer Science and Technology, Guizhou University, Guiyang 550025, China;
    2. Guizhou Engineering Laboratory for Advance Computing and Medical Information Service, Guiyang 550025, China;
    3. National Astronomical Observatories, Chinese Academy of Sciences, Beijing 100101, China)
  • Received:2020-02-06 Online:2020-05-20 Published:2020-05-21

摘要: 在资源受限条件下,根据数据挖掘任务在执行过程中实时产生的资源和任务状态来准确地预测任务执行时间是非常重要的。为有效地使用时间序列数据实现准确预测,提出一种降载策略来确定预测的切入点和数据处理方案。该策略使用动态时间规整(Dynamic Time Warping, DTW)距离度量子序列与整个序列之间相似度的变化以确定用于预测的数据,然后利用小波变换计算小波系数并提取小波系数的能量值作为预测的特征,最后预测任务执行时间。实验结果表明,该方法提取的特征信息包含原序列较多信息,在预测任务执行时间方面具有较高的准确性。

关键词: 降载, 小波变换, 特征提取, 任务执行时间预测

Abstract: In resource constraints condition, it is very important to make accurate predictions of the task execution time based on time-series resource and task status generated in real-time during task execution. In order to use time-series data effectively to realize accurate prediction, a load shedding strategy is proposed to determine the time points of prediction and data processing scheme. This strategy uses dynamic time warping (DTW) distance to measure the variation of similarity between subsequences and entire sequences and determine the data used for prediction. Then we use wavelet transform to calculate the wavelet coefficients of the time-series and extract the energy value of wavelet coefficients as the features of prediction. After that, we conduct the prediction for task execution time. Experiments show that the features extracted by this method contain most information than the entire sequence and result in high accuracy in predicting the task execution time.

Key words: load shedding, wavelet transform, feature extraction, task execution time prediction

中图分类号: