Multivariate Time Series Regression on Seismic Data Using Adaptive Residual CNNs
Autor(en): | van den Hoogen, Jurgen O.D. Bloemheuvel, Stefan D. Atzmueller, Martin |
Herausgeber: | Reuss, P. Eisenstadt, V. Schonborn, J. Schafer, J. |
Stichwörter: | Convolutional neural network; Convolutional Neural Networks; Deep Learning; Input layers; Large dataset; Large-scale datasets; Learning systems; Long short-term memory; Machine-learning; Multivariate time series; Regression analysis; Residual network; Residual Networks; Seismic datas; Seismology; Sensors; Time series; Time series analysis; Time Series Regression; Time-series data; Time-series regression | Erscheinungsdatum: | 2022 | Herausgeber: | CEUR-WS | Journal: | CEUR Workshop Proceedings | Volumen: | 3341 | Startseite: | 79 – 90 | Zusammenfassung: | Developments in Machine Learning and more specifically Deep Learning have boosted the analysis of large-scale datasets. These methods for automated learning are particularly useful for complex data originating from sensors, i. e., multivariate time series data. In this work, we present an adaptive residual CNN (ARes-CNN) that is able to process such multivariate time series data. The model utilizes an adaptive input layer that separately processes every time series (e. g., channels or sensors), learning their key individual characteristics. Furthermore, the model applies stacked residual learning throughout each layer. We compare ARes-CNN with traditional Machine Learning applications, as well as a CNN and LSTM developed for the task at hand. The models' performance are compared on two datasets retrieved from sensors in a network of seismic stations located in Italy. Across all experiments, ARes-CNN reports a performance increase of 17% (MSE) on average compared to the best performing baseline. Therefore, processing the channels independently (by employing adaptive input layers), together with residual learning are fit for multivariate time series regression, e. g., for analysis of seismic sensor data. © 2022 Copyright for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0). |
Beschreibung: | Cited by: 0; Conference name: 2022 Learning, Knowledge, Data, Analysis, LWDA 2022 - Workshops: Special Interest Group on Knowledge Management (FGWM), Knowledge Discovery, Data Mining, and Machine Learning (FGKD) and Special Interest Group on Database Systems (FGDB); Conference date: 5 October 2022 through 7 October 2022; Conference code: 186661 |
ISSN: | 1613-0073 | Externe URL: | https://www.scopus.com/inward/record.uri?eid=2-s2.0-85148637437&partnerID=40&md5=b419f65c3f26315f6e1b3e2bcb69775e |
Show full item record