Modeling Control Signals for Reconstruction-based Time Series Anomaly Detection

Automated time series anomaly detection methods can provide insights while reducing the load placed on human experts in a variety of settings. Machine-generated signals, such as those produced by sensors, often contains control signals in addition to the target observation signal. These signals may provide additional insight about the normal vs. abnormal properties of the observation signal. Despite this fact, even recent anomaly detection methods using deep learning give limited consideration to the relationship between observation and control signals, often failing to handle the control signal at all. This work proposes pre-processing, modeling, and evaluation methods for multivariate, heterogeneous time series to examine how using information from the control signal can improve anomaly detection. We develop a deep learning reconstruction-based pipeline and test its performance on the NASA Soil Moisture Active Passive (SMAP) satellite and the Mars Science Laboratory (MSL) Rover, which contains heterogeneous sensing data from exploratory missions. The pipeline follows the Sintel machine learning framework and is accessible through the Meissa library, which builds on the capabilities of the open-source library Orion for end-to-end unsupervised time series anomaly detection pipelines.

Name thesis:degree_name

Master

Department dc:contributor.department

Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science

Grantor dc:publisher

Massachusetts Institute of Technology

Year dc:date.issued

2024

Author dc:creator

• Song, Grace Y.

Advisor dc:contributor.advisor

• Veeramachaneni, Kalyan

Statement dc:rights

• Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0)
• Copyright retained by author(s)

Licence dc:rights.uri

• https://creativecommons.org/licenses/by-nc-nd/4.0/

Handle dc:identifier.uri

https://hdl.handle.net/1721.1/156789

OAI identifier oai:identifier

oai:dspace.mit.edu:1721.1/156789