Résumé
Many phenomena can be observed and organized as a sequence of observations on a timeline that can be modeled as a time series. A relevant area that is being explored in time series analysis is pattern discovery. Patterns are subsequences of time series related to some special properties or behaviors. A particular pattern that occurs a significant number of times in time series is called motif. Several important phenomena of a time series present different behaviors when observed at points in space (for example, series collected by sensors and IoT) and are best modeled as space-time series. Each time series is associated with a position in space. A space-time pattern is a sequence of events that are limited in space and time. Finding patterns that are frequent and restricted in space and time can allow us to understand how a phenomenon occurs.Several works have been developed to identify motifs in time series. However, studies that address spatiotemporal data techniques have not been identified. In this thesis, we compare different approaches to identifying motifs in time series with their main differences. We propose two methods for automatically identifying space-time restricted motifs in space-time series, the CSA and CSTMP. We experimentally compare the pro- posed methods with two other alternative methods: the Matrix Profile technique and the ensemble of the Matrix Profile and DBScan techniques. Our results show that CSA and CSTMP are innovative and obtain results that outperform the state-of-the-art techniques for time series and their adaptations for space-time series, being evaluated in two datasets.