利用coherence分析於多變量時間序列資料之分群

時間序列資料經由頻譜轉換後，可以使資料有更好的性質應用於資料分群中，然而目前資料分群的方法雖已廣泛的應用於單變量之時間序列資料中，對於多變量的時間序列資料的分群卻較少著墨。本研究提出以經由頻譜轉換之Coherence，作為分群依據的分群方法，期待能找到一個在多維度下適當的Coherence加權，並模擬多維度的時間序列資料，在考慮不同變數(資料維度、樣本數、分群方法…等)，觀察分群結果。研究結果發現，在模擬資料的部分，當資料有加入干擾的變數或是非定態的情形產生時，Cohrence的分群方法仍有不錯的表現。而在真實資料分群的部分，也能對比真實世界的情形提供資訊，從中得到理想的建議。

關鍵字

多變量時間序列分群；頻譜轉換； coherence

並列摘要

The clustering analysis of time series data have been applied in many fields extensively. The application of coherence method in the spectral clustering analysis can overcome the problems among the traditional distance measuring clustering; hence derives an admired clustering result. However, most of the researches in coherence clustering are focusing in the univariate setup. In this study, we propose a multivariate coherence clustering algorithm. The performance will be gauged through real data analysis and simulations under different conditions and parameters. We look forward to have a good result of multivariate time series clustering.

並列關鍵字

multivariate time series data clustering ； coherence

參考文獻

[1] Elizabeth Ann Maharaj, Pierpaolo D'Urso, “A coherence-based approach for the pattern recognition of time series”, Physica A: Statistical Mechanics and its Applications, 389(2010), 3516-3537

[2] Peter D.Welch, “The use of fast Fourier transform for the estimation of power spectra: A method based on time averaging over short, modified periodograms”, IEEE Transactions on Audio and Eletroacoustics, 15(1967), 70-73

[3] G. Clifford Carter, Charles H. Knapp, Albert H. Nuttal, “Estimation of magnitude-squared coherence function via segment overlapped fast fourier transform precessing”, IEEE Transactions on Audio and Eletroacoustics, No.4(1973), 1100-1117

[5] Leonard Kaufman, Peter J. Rousseeuw, “Finding groups in data: An introduction to cluster analysis”, (1990)

[6] Lawrence Huber, Phipps Arabie, “Comparing partitions”, Journal of Classification, No.1(1985),193-218

國際替代計量

利用coherence分析於多變量時間序列資料之分群

未授權

主題瀏覽