Publikation
Time Series Similarity Search for Streaming Data in Distributed Systems
Ariane Ziehn; Marcela Charfuelan Oliva; Holmer Hemsen; Volker Markl
In: Proceedings of the Workshops of the EDBT/ICDT 2019 Joint Conference (EDBT/ICDT 2019). Data Analytics Solutions for Real-Life Applications (DARLI-AP-2019), located at EDBT/ICDT 2019 Joint Conference, March 26-29, Lissabon, Portugal, CEUR-WS, 2019.
Zusammenfassung
In this paper, we propose a practical study and demonstration of time series similarity search in modern distributed data processing platforms for stream data. After an intensive literature review, we implement a flexible similarity search application in Apache Flink, which includes the most commonly used distance measurements: Euclidean distance and Dynamic Time Warping. For efficient and accurate similarity search we evaluate normalization and pruning techniques developed for single machine processing and demonstrate that they can be adapted and leveraged for those distributed platforms. Our final implementation is capable of monitoring many time series in real-time and parallel. Further, we demonstrate that the number of required parameters can be reduced and optimally derived from data properties. We evaluate our application by comparing its performance with electrocardiogram data on a cluster with several nodes. We reach average response times of less than 0,1 ms for windows of 2 s of data, which allow fast reactions on matching sequences.
Projekte
- BigMedilytics - Big Data for Medical Analytics
- Daystream - Data Analytics and AI for Secure, Trusted, and Reliable Mobility (Datenanalytik und KI für sichere und zuverlässige Mobilität)