Detecting time series anomalies using hybrid methods applied to Radon signals recorded in caves for possible correlation
- PDF / 1,528,478 Bytes
- 16 Pages / 439.37 x 666.142 pts Page_size
- 78 Downloads / 208 Views
Detecting time series anomalies using hybrid methods applied to Radon signals recorded in caves for possible correlation with earthquakes F. Ambrosino1,2 · L. Thinová2 · M. Briestenský3 · S. Šebela4 · C. Sabbarese1 Received: 10 July 2019 / Accepted: 19 May 2020 © Akadémiai Kiadó 2020
Abstract Anomalies in the Radon activity concentration time series recorded in five European caves (Czech Republic, Slovakia, Slovenia) are detected using three hybrid methods: (1) multiple linear regression and autoregressive integrated moving average statistical methods, (2) Empirical Mode Decomposition with Support Vector Regression techniques and (3) the Singular Spectrum Analysis composed with a predicting methodology. Results coming from the three methods are compared and the best hybrid method is selected based on statistical evaluation criteria of the uncertainty. Radon anomalies occur ± 30 days from earthquake occurrence, selected according to the Dobrovolsky’s earthquake preparation zone formula and to seismic events (with magnitude ≥ 4) occurred in the neighboring European Countries to the monitoring caves. The anomalies detection furnishes results consistent across the used methodologies, as proven by the calculation of a statistical parameter that search the presence of anomalies coming from the hybrid methods within ± 30 days from earthquake event. Keywords Time series analysis · Hybrid method · Anomaly detection · Radon as tracer · Earthquake
* F. Ambrosino [email protected] 1
Department of Mathematics and Physics, University of Campania “Luigi Vanvitelli”, Viale Lincoln 5, 81100 Caserta, Italy
2
Faculty of Nuclear Sciences and Physical Engineering, Czech Technical University in Prague, Břehová 7, 11519 Prague 1, Czechia
3
Institute of Rock Structure and Mechanics, Czech Academy of Sciences, V Holešovičkách 41, 18209 Prague 8, Czechia
4
Karst Research Institute, Slovenia Research Centre of the Academy of Sciences and Arts, Titov trg 2, 6230 Postojna, Slovenia
13
Vol.:(0123456789)
Acta Geodaetica et Geophysica
1 Introduction 1.1 Anomaly detection methods An increasing number of time series analysis methods have been implemented by many researchers for different purposes over the past few decades (Ambrosino et al. 2020c; Bashir and El-Hawary 2009; Cadenas and Rivera 2010; Kavitha and Raghukanth 2016; Lin 2013; Sabbarese et al. 2017b; Stathopoulos et al. 2013). Examples of the most utilized methods are: ARIMA-autoregressive integrated moving average, MLR-multiple linear regression, ANN-neural networks, SSA-singular spectrum analysis, EMD-empirical mode decomposition, SVR-support vector regression. Deficiencies and characteristics are present in every method, i.e. based on the management of the non-linear problem (Cadenas and Rivera 2010), on the requirement of many algorithm parameters (Bashir and El-Hawary 2009) and on the consumption of computing resources (Stathopoulos et al. 2013). So, a single traditional algorithm cannot fully discover and capture the time series features due t
Data Loading...
