Home // CLOUD COMPUTING 2025, The Sixteenth International Conference on Cloud Computing, GRIDs, and Virtualization // View article

A Transformer-Based Framework for Anomaly Detection in Multivariate Time Series

Authors:
Fabian Folger
Murad Hachani
Philipp Fuxen
Julian Graf
Sebastian Fischer
Rudolf Hackenberg

Keywords: AI; Transformer; Time Series; Anomaly Detection; ECU; Temporal Aggregation

Abstract:
This paper introduces a comprehensive Transformer-based architecture for anomaly detection in multivariate time series. Using self-attention, the framework efficiently processes high-dimensional sensor data without extensive feature engineering, enabling early detection of unusual patterns to prevent critical system failures. In a subsequent laboratory setup, the framework will be applied using fuzzing techniques to induce anomalies in an Electronic Control Unit, while monitoring side channels, such as temperature, voltage, and Controller Area Network messages. The overall structure of the architecture, as well as the necessary pre-processing steps, such as temporal aggregation and classification up to the optimization of the hyperparameters of the model, are presented. The evaluation of the model architecture with the postulated restrictions shows that the model handles anomaly scenarios in the dataset robustly. It is necessary to evaluate the extent to which the model can be used in practical applications in areas, such as cloud environments or the industrial Internet of Things. Overall, the results highlight the potential of Transformer models for the automated and reliable monitoring of complex time series data for deviations.

Pages: 52 to 57

Copyright: Copyright (c) IARIA, 2025

Publication date: April 6, 2025

Published in: conference

ISSN: 2308-4294

ISBN: 978-1-68558-258-6

Location: Valencia, Spain

Dates: from April 6, 2025 to April 10, 2025