Interactive discovery of sequential patterns in time series of wind data

Abstract

Wind speed and direction vary over space and time due to the interactions between different pressures and temperature gradients within the atmospheric layers. Near the earth's surface, these interactions are modulated by topography and artificial structures. Hence, characterizing wind behaviour over large areas and long periods is a complex but essential task for various energy-related applications. In this study, we present a novel approach to discover wind patterns by integrating sequential pattern mining and interactive visualization techniques. The approach relies on the use of the Linear time Closed pattern Miner sequence algorithm in conjunction with a time sliding window that allows the discovery of all sequential patterns present in the data. These patterns are then visualized using integrated 2D and 3D coordinated multiple views and visually explored to gain insight into the characteristics of the wind from a spatial, temporal and attribute (type of wind pattern) point of view. This proposed approach is used to analyse 10 years of hourly wind speed and direction data for 29 weather stations in the Netherlands. The results show that there are 15 main sequential patterns in the data. The spatial task shows that weather stations located in the same region do not necessarily experience similar wind pattern. For within the selected time interval, similar wind patterns can be observed in different stations and in the same station at different times of occurrence. The attribute task discovered that the repetitive occurrences of chosen pattern indicate as regular wind behaviour at different weather stations that persisted continuously over time. The results of these tasks show that the proposed interactive discovery facilitates the understanding of wind dynamics in space and time.ARTICLE HISTORY

Discretized sequential pattern mining for behaviour classification

“…Tseng and Lee (2009) propose a discretization method for use with SPM that assigns adjacent data points to one of 20 levels according to their differences. Yusof et al (2016) perform SPM on wind data, for which speed and direction are discretized into 7 and 8 different levels, respectively. Finally, Das et al (1998) propose an adjustable approach for discretization for the purpose of rule discovery, where data segment windows of variable size can be selected for discretization, which can then be conducted at various levels of granularity.…”

Geospatial big data and cartography: research challenges and opportunities for making maps that matter

“…Examples include the analysis of changing temperature over the last 30 years at one particular weather station (Wu, Zurita-Milla, & Kraak, 2015). Another example could be the changing wind pattern over multiple years at a particular location observed at different heights (Yusof, Zurita-Milla, Kraak, & Retsios, 2016). For both examples, the complexity will increase if we consider multiple locations with observations.…”

A new approach for spatio-temporal data mining

“…L'exploration des données spatiales comme des données temporelles peut donc être ramenée à une visualisation 2D animée ou statique [5]. Cela peut donc permettre d'explorer des ensembles de données denses et complexes [13,14].…”

“…Leur visualisation utilise généralement des rendus 2D, des visualisations animées [5] ou encore le concept de cube spatio-temporel [1,14] (cf. Figure 1).…”