時間序列分析的圖案計量

本論文提出一個新的序列圖案架構，可以正確並快速的計算出在時間序列資料中圖案出現的機率。此架構利用自動機描述出現圖案的計量，並將這個自動機嵌入一個馬可夫鏈，而圖案計量的分佈即可由馬可夫鏈的分佈求得。這個架構可以用來分析連續或離散的序列資料，只要圖案是出現於資料模型中的一個馬可夫來源。透過這個新方法，可以簡單並有效率的取得圖案計量的共同分佈。這個新方法的應用範圍極廣，本論文中將示範如何將之應用於時間序列中變異點的估計。

關鍵字

連列與圖案；圖案計量；時間序列分析；馬可夫鏈；隱含馬可夫模型；馬可夫鏈嵌入；有限自動機；形式語言；變異點；非平穩型時間序列

並列摘要

This thesis introduces a new pattern statistics framework, which enables exact and efficient calculation of probabilities of pattern occurrences in sequence data. Statistics of pattern occurrences in data are formulated in terms of finite automata state transitions embedded into a Markov chain. This enables the analysis of continuous or discrete sequence data where the underlying generation process is governed by a Markov source, and where occurrences of specific patterns in the Markov state sequence is of interest. Through this new methodology, the full joint distribution of pattern statistics can be obtained in a conceptually simple and computationally efficient way. This new methodology can be adopted for many applications, and is here applied to change point estimation problems as an example.

並列關鍵字

Runs and patterns ； pattern statistics ； time series analysis ； Markov chains ； hidden Markov models ； Markov chain embedding ； deterministic finite automata ； regular languages ； change points ； non-stationarity time series

參考文獻

James C. Fu and Yung-Ming Chang. On ordered series and later waiting time

bibliographic search. Communications of the ACM, 18(6):333{340, Jun 1975.

doi: 10.1145/360825.360855.

Sigeo Aki. Discrete distributions of order k on a binary sequence. Annals

of the Institute of Statistical Mathematics, 37(2):205{224, Dec 1985. doi:

國際替代計量

時間序列分析的圖案計量

全文下載

主題瀏覽