ENBIS-17 in Naples
9 – 14 September 2017; Naples (Italy)
Abstract submission: 21 November 2016 – 10 May 2017
Anomaly Detection in Maritime Data Streams
12 September 2017, 16:00 – 16:20
- Submitted by
- Ingrid Kristine Glad
- Ingrid Kristine Glad (University of Oslo), Andreas Brandsæter (University of Oslo/DNV-GL), Martin Tveten (University of Oslo)
- In the shipping industry, there is an increasing interest in monitoring ship operations through sensor data continuously streamed from vessels to shore. Data streams represent very diverse aspects of the ship´s state and performance, while the ships operate in several modes and under various climatic conditions.
In order to be fed into (sequential) tests for anomaly detection, such data streams are first taken through a data driven, nonparametric method for signal reconstruction known as Auto Associative Kernel Regression (AAKR). Streams of residuals (observed - reconstructed) are then monitored successively. We compare the classical univariate Sequential Probability Ratio Test (SPRT) with more recent extensions to multivariate (and possibly high dimensional) procedures for sequential detection of changes in mean and/or variance and covariance. These sequential tests are also compared to other change point detection procedures, with the aim of detecting changes as soon as possible, while controlling the amount of false alarms.
Return to programme