Real-Time Anomaly Detection in Underwater Gliders: Anomaly Detection Algorithm

22 May 2024


(1) Ruochu Yang;

(2) Chad Lembke;

(3) Fumin Zhang;

(4) Catherine Edwards.

Abstract and Intro

Anomaly Detection Algorithm

Experimental Evaluation

Conclusion and References


The pipeline of anomaly detection and false alarm elimination is shown in Fig. 1. By generating the glider speed estimate, the algorithm assumes no anomaly if the estimate is within the normal speed range. Otherwise, the glider may be encountering issues. The flow speed estimate is checked against the glider-estimated flow speed to circumvent any false alarm.

We model the glider dynamics as follows:

where θ is the unknown parameter to be estimated,

Fig. 1: Pipeline of anomaly detection and false alarm elimination. Three steps: 1) generate glider speed estimate and flow speed estimate based on DBD or SBD data; 2) compare glider speed to detect anomalies; 3) compare flow speed for false alarm.

This paper is available on arxiv under CC 4.0 license.