北京邮电大学学报

  • EI核心期刊

北京邮电大学学报 ›› 2021, Vol. 44 ›› Issue (6): 96-102,133.doi: 10.13190/j.jbupt.2021-045

• 论文 • 上一篇    下一篇

敏捷化AIOps框架及运维数据质量评估方法

吴振宇, 施畅   

  1. 北京邮电大学 教育部信息网络工程研究中心, 北京 100010
  • 收稿日期:2021-04-09 出版日期:2021-12-28 发布日期:2021-12-28
  • 作者简介:吴振宇(1986—),男,副教授,E-mail:shower0512@bupt.edu.cn.
  • 基金资助:
    广东省重点领域研发计划项目(2020BO101130013);2020年工业互联网创新发展工程项目(2020GXb001)

Agile AIOps Framework and Maintenance Data Quality Assessment Method

WU Zhen-yu, SHI Chang   

  1. Engineering Research Center for Information Network(Ministry of Education), Beijing University of Posts and Telecommunications, Beijing 100010, China
  • Received:2021-04-09 Online:2021-12-28 Published:2021-12-28

摘要: 敏捷化智能运维(AIOps)框架将模型构建提前至测试阶段,利用该阶段产生的监控数据代替线上采集的数据以训练AIOps模型,进而实现智能运维的早开发与早使用. 运维数据质量评估方法通过最大均值差异度量方式,分别在健康评估与故障诊断运维场景下对训练数据分别进行趋势性、阶段性、可检测性及可诊断性评估,以预估数据对模型的适用性. 基于华为技术有限公司提供的测试环境设置测试用例并构建实验数据集,在该数据集上的实验结果验证了敏捷化AIOps框架的可行性及数据质量评估方法的有效性.

关键词: 智能运维, 敏捷化框架, 数据质量评估, 最大均值差异

Abstract: An agile artificial intelligence for information technology operations(AIOps)framework and maintenance data quality assessment method are proposed. The agile AIOps framework advances the model construction stage to the test stage, and uses the monitoring data generated during the test stage to replace the data collected online to train the model, thereby realizing the early development and early use of intelligent operation. The maintenance data quality assessment method is based on the maximum mean discrepancy to evaluate the trend, stage, detectability, and diagnosability of training data for health assessment and fault diagnosis, so as to estimate the applicability of the data to the model. Based on the test environment provided by Huawei, the test cases are set up and the experimental data set is constructed. The experimental results on the data set verify the feasibility of the agile AIOps framework and the effectiveness of the data quality assessment method.

Key words: artificial intelligence for information technology operations, agile framework, data quality assessment, maximum mean discrepancy

中图分类号: