Time series anomaly detection for gravitational-wave detectors based on the Hilbert–Huang transform

Abstract

We present a new event trigger generator based on the Hilbert-Huang transform, named EtaGen ( Gen). It decomposes time-series data into several adaptive modes without imposing a priori bases on the data. The adaptive modes are used to find transients (excesses) in the background noises. A clustering algorithm is used to gather excesses corresponding to a single event and to reconstruct its waveform. The performance of EtaGen is evaluated by how many injections are found in the LIGO simulated data. EtaGen is viable as an event trigger generator when compared directly with the performance of Omicron, which is currently the best event trigger generator used in the LIGO Scientific Collaboration and Virgo Collaboration.

Anomaly Detection in Connected and Autonomous Vehicle Trajectories Using LSTM Autoencoder and Gaussian Mixture Model

“…Most statistical anomaly detection methods struggle to model volatility and long-term trends in time-series data. This is because they often depend on methods that work better on simpler datasets and fail to capture the fluctuations of variables with little correlations in the system [28][29][30]. In recent years, deep neural networks have become very popular and are used in many advanced models.…”

Identifying multichannel coherent couplings and causal relationships in gravitational wave detectors

“…Studies of finding coherence between the strain channel and auxiliary channels of the GW detectors have been extensively performed in LIGO-Virgo collaborations, in the context of continuous GW and stochastic GW background searches [11][12][13][14][15][16]. Many efforts for identifying and vetoing transient noises have been made so far and are widely utilized for GW data analysis, such as a computation of significance called hierarchical veto (Hveto) [17] and Used percentage veto (UPV) [18], noises associated with longduration transients [19,20], Q-transform based trigger generator [21], the Hilbert-Huang transform-based method [22], the bicoherence method [23], linear regressions [24], machine learning algorithms [25], and so on. These methods cover classification and a vetoing method as well as identification of the influences between the GW strain channel and auxiliary channels monitoring the environmental and/or instrumental status.…”