Hi Augusto,

You may want to look at a technique called "Empirical Orthogonal Functions (EOF)": they can represent, in an optimal way, the variance of time series but they also automatically generate spatial patterns that may be of interest too. A search on the Internet for these keywords just generated 137,000 results... Cheers, Michel.

Hi Augusto, I am thinking about using correlation, or maybe especially distance correlation (http://en.wikipedia.org/wiki/Distance_correlation).

Hi Augusto,

Common ways of comparing time series is to compare their values using a given similarity and distance measure (e.g., X. Wang, A. Mueen, H. Ding, G. Trajcevski, P. Scheuermann, and E. Keogh, “Experimental comparison of representation methods and distance measures for time series data,” Data Mining Knowl. Discovery, vol. 26, p. 275, 2013)

There are very many measures of dependency that can be measured -- things from simple linear correlations through to things like transfer entropy, Granger causality, and other information theoretic-based measures.

I also recently proposed a method where you compare a set of useful features summarizing structure in time series, and learn the best such features for a given application (Fulcher, B. D., & Jones, N. S. (2014). Highly comparative feature-based time-series classification. IEEE Transactions on Knowledge and Data Engineering, 26(12), 3026–3037. doi:10.1109/TKDE.2014.2316504)

Hope this helps :)

Hi Augusto,

suggestions:

Published:

Measuring the distance between time series

Richard Moeckel, Brad Murrayb, 

doi:10.1016/S0167-2789(96)00154-6

Unpublished:

http://cran.r-project.org/web/packages/TSdist/vignettes/TSdist.pdf

http://www.cs.ucr.edu/~eamonn/Complexity-Invariant%20Distance%20Measure.pdf

http://wwweb.eecs.umich.edu/db/files/sigmod07timeseries.pdf

http://www.iiia.csic.es/files/pdfs/serraarcos_iccbr12.pdf

I suggest Higuchi method