0


时间序列数据马尔可夫模型的序列假设检验

Sequential Hypothesis Tests for Markov Models of Time-Series Data
课程网址: http://videolectures.net/kdd2016_virani_markov_models/  
主讲教师: Nurali Virani
开课单位: 宾夕法尼亚州立大学
开课时间: 2016-11-07
课程语种: 英语
中文简介:
本文提出了时间序列数据马尔可夫模型的序贯假设检验的新结果。特别地,开发了一种使用符号动力学概念推导的马尔可夫模型序列假设检验技术。这些模型是通过离散动力系统的相空间建立的,系统动力学近似为离散状态空间上的有限记忆马尔可夫链。在二元假设检验的前提下,提出了马尔可夫模型对数似然比统计量的序列更新规则,并分析了该统计量的随机演化规律。所提出的技术允许我们选择探测器性能的下界,并保证测试将在有限时间内终止。该研究的动机是物理系统的时间关键检测问题,其中训练的时间模型和在运行过程中需要大量流数据的快速可靠的决策。首先通过一个仿真实例说明了所提出的技术。此外,在实验室规模的涡流稳定燃烧室获得的压力时间序列数据上进行了测试,其中一些控制协议被用于诱导不稳定。提出的框架用于检测和估计燃烧过程中不稳定的开始。我们将其与最大似然分类器的性能进行了比较,并表明所提出的技术使用较少的观测量提供了可靠的不稳定性检测。
课程简介: This paper presents new results on sequential hypothesis tests for Markov models of time series data. In particular, a technique for sequential hypothesis testing for Markov models inferred using concepts of symbolic dynamics is developed. These models are created by discretizing the phase space of a dynamical system and the system dynamics is approximated as a finite memory Markov chain on the discrete state-space. We present sequential update rules for log-likelihood ratio statistic of Markov models under the setting of binary hypothesis testing and analyze the stochastic evolution of this statistic. The proposed technique allows us to choose a lower bound on the performance of the detector and guarantees that the test will terminate in finite time. The study is motivated by time-critical detection problems with physical systems, where a temporal model is trained and a fast reliable decision with large volumes of streaming data is desired during operation. The proposed technique is first illustrated through a simulation example. Furthermore, the ideas are tested on pressure time-series data obtained from a laboratory-scale swirl stabilized combustor, where some controlled protocols are used to induce instability. The proposed framework is used to detect and estimate onset of instability during combustion. We compare the performance with maximum-likelihood classifier and show that the proposed technique gives reliable detection of instability using fewer observations.
关 键 词: 时间序列; 仿真实例; 马尔可夫模型
课程来源: 视频讲座网
数据采集: 2022-11-03:chenjy
最后编审: 2023-05-13:chenjy
阅读次数: 33