Markov chain embedded temporal pattern analysis
碩士 === 國立東華大學 === 應用數學系 === 98 === This work explores Markov chain embedded temporal pattern analysis. The main task includes temporal pattern segmentation, recursive nonlinear kernel extraction, temporal pattern identification, hidden state allocation and transition probability estimation. A Markov...
| Main Authors: | , |
|---|---|
| Other Authors: | |
| Format: | Others |
| Language: | en_US |
| Published: |
2010
|
| Online Access: | http://ndltd.ncl.edu.tw/handle/89819967905837234132 |
| Summary: | 碩士 === 國立東華大學 === 應用數學系 === 98 === This work explores Markov chain embedded temporal pattern analysis. The main task includes temporal pattern segmentation, recursive nonlinear kernel extraction, temporal pattern identification, hidden state allocation and transition probability estimation. A Markov chain of pairwise Gaussian mixture (PGM) models is employed to characterize given time series. A PGM model that consists of multiple pairs of normal random variables is employed to emulate formation of given paired training data. Auto-regressive sampling translates a segment of temporal pattern to a set of paired data whose fitting to a PGM model induces the task of recursive neural function approximation which is resolved by supervised learning of a network of normalized RBFs (Radial basis functions). Fitting a PGM model involves extracting nonlinear recursive structures underlying a segment of temporal pattern for time series identification. A novel systematic approach is proposed for reconstruction of a Markov chain of PGM models. The proposed Markov chain embedded temporal pattern analysis is applied to synthesize long term chaotic time series and real world signals.
|
|---|