A data-driven trimming method in time series clustering with applications to the study of random sea waves

P. C. Álvarez Esteban, J. Ortega, C. Euán

In this work we tackle the problem of studying the different stationary periods of random sea waves measured in a fixed point of the sea through the use of a times series clustering algorithm. The two key points of this algorithm are the use of the total variation distance between the normalized spectra of the time series and the use of what is called impartial trimmings. Total variation distance is used then as a measure of the similarity between time series, and the focus is on the energy distribution more than on the total energy present as we normalize the spectra. On the other hand, trimming methodology is introduced to avoid the effect of transition periods on the clustering process. We present simulation studies to validate the proposed method as well as an application to real data.

Palabras clave: Clustering total variation distance trimmed k-means spectral analysis sea state stationary periods.

G. Villarino, R. Cintas del Río, D. Gómez González, J. T. Rodriguez González

C. Rendón Aguirre, D. Peña, F. J. Prieto Fernández

R. Morales Arsenal, J. J. Núñez Velázquez