Figure 6 | Scientific Reports

Figure 6

From: Model-free detection of unique events in time series

Figure 6

Detection of the GW150914 event on LIGO open data with TOF and LOF and discord. (A) Strain time series (black) from Hanford detector around GW150914 event (grey vertical line) with TOF (orange dots), LOF (blue plus) and discord (red x) detections. TOF score values (B), LOF scores (C) and matrix profile scores (D) are mapped to the time series (orange, blue and red colors respectively), the strongest colors show the detected event around 0 s. (E) The Q-transform of the event shows a rapidly increasing frequency bump in the power spectra right before the merger event (grey). The grey horizontal dashed lines show the lower (50 Hz) and upper (300 Hz) cutoff frequencies of the bandpass filter, which was applied on the time series as a preprocessing step before anomaly detection. (F) Filtered strain data at 0.1 s neighborhood around the event. TOF, LOF, and discord detection algorithms detected the merger event with different sensitivity. LOF detected more points of the event, while TOF found the period which has the highest power in the power spectra, and a discord was detected at the end of the event. (\(E_{\mathrm{TOF}}=6\), \(\tau _{\mathrm{TOF}}=1.953\) ms, \(k_{\mathrm{TOF}}=12\), \(M_{\mathrm{TOF}}=146.484\) ms, \(w=7\); \(E_{\mathrm{LOF}}=11\), \(\tau _{\mathrm{LOF}}=1.953\) ms, \(k_{\mathrm{LOF}}=100\), threshold\(=0.5\)%, \(M_{discord}=146.484\,{\text{ms}}\)).

Back to article page