本文为捷克马萨里克大学（作者：Bc. Pavol Loffay）的硕士论文，共67页。

本硕士论文的目标是开发一个开放源码的监控与管理平台Hawkular。该模块应为采集的时间序列数据提供实时预测能力。所提出的解决方案应该具有计算效率，更重要的是，它应该为给定的度量自动选择最佳的时间序列模型。换句话说，它应该在没有分析师交互的情况下工作。本硕士论文从时间序列理论入手，分析了时间序列预测的各种方法。讨论了哪些模型最适合Hawkular需求。实现章节主要关注的最重要部分，如时间序列模型、自动预测和与Hawkular的集成。最后一章对实现模型进行了评价。将模型与统计语言R的备选方案进行比较，结果表明，从预测精度的角度来看，所实现的模型与R的备选方案相似。

The goal of the master’s thesis is todevelop a module for an open source monitoring and management platformHawkular. The module should provide real time predictive capabilities forcollected time series data. Proposed solution should be computationallyefficient and more importantly, it should autonomously select the best timeseries model for given metric. In other words it should work without analyst’sinteraction. The master’s thesis starts with time series theory and analysis ofvarious approaches for time series forecasting. It discusses which models arebest for Hawkular requirements. The implementation chapter focuses on the mostimportant parts like time series models, automatic forecasters and integrationinto Hawkular. The last chapter conducts evaluation of the implemented models.Models are benchmarked against its alternatives from statistical language R.The results show that implemented models are similar to R alternatives from theprediction accuracy perspective.

1. 引言

2. 时间序列模型

3. 分析预测过程

4. 已有的解决方案

5. 设计与实现

6. 实现模型的评估

7. 结论

更多精彩文章请关注公众号：