Percorrer por autor "Cardoso, Margarida"
A mostrar 1 - 2 de 2
Resultados por página
Opções de ordenação
- An application of time series clustering using a combined distancePublication . Martins, Ana; Vaz, Daniel; Silva, Tiago; Cardoso, Margarida; Carvalho, Alda; Loja, AméliaClustering time series aims at uncovering diverse longitudinal patterns. In this study we analyse time series of various parameters collected atop wind turbines in a wind farm in Portugal. Considering wind data in clustering may reveal differences in operation between neighbouring turbines, due to their position relative to one another and to terrain features. In this work, we use an approach to wind speed time series clustering based on a convex combination of distance measures between time series. For visualizing the resulting groups, we propose a graphical representation, Distance Matrix, which is quite more informative than the classical Multidimensional Scaling (MDS) map. This representation allows for quick comparisons between pairs of turbines for (dis)similarities. Our approach provides distinct insights regarding the differences between time series, emphasizing differences in values (Euclidean distance), trends (Pearson-based distance), and cyclical behaviours (Euclidean distance between periodograms and/or autocorrelation structures). In most cases, we found two groups, which were not always coincident with the geographical groups, but the proposed approach could also find the rationale behind the clusters that were formed. The results obtained may help identifying undesirable aerodynamic loads that the blades of a particular wind turbine may be subjected to, thereby shortening its time in-service.
- Clustering of wind speed time series as a tool for wind farm diagnosisPublication . Martins, Ana; Vaz, Daniel; Silva, Tiago; Cardoso, Margarida; Carvalho, AldaIn several industrial fields, environmental and operational data are acquired with numerous purposes, potentially generating a huge quantity of data containing valuable information for management actions. This work proposes a methodology for clustering time series based on the K-medoids algorithm using a convex combination of different time series correlation metrics, the COMB distance. The multidimensional scaling procedure is used to enhance the visualization of the clustering results, and a matrix plot display is proposed as an efficient visualization tool to interpret the COMB distance components. This is a general-purpose methodology that is intended to ease time series interpretation; however, due to the relevance of the field, this study explores the clustering of time series judiciously collected from data of a wind farm located on a complex terrain. Using the COMB distance for wind speed time bands, clustering exposes operational similarities and dissimilarities among neighboring turbines which are influenced by the turbines’ relative positions and terrain features and regarding the direction of oncoming wind. In a significant number of cases, clustering does not coincide with the natural geographic grouping of the turbines. A novel representation of the contributing distances—the COMB distance matrix plot—provides a quick way to compare pairs of time bands (turbines) regarding various features.