D3: ID.10671 Seismic Anomalies Detection
Wednesday, 06/Jul/2016:
9:00am - 10:00am

Session Chair: Nico Sneeuw
Session Chair: Mingsheng Liao
Workshop: Terrain & Geoid Measurement
Location: A-Li Mountain Room, 5.5 Floor, Junyi Dynasty Hotel

Oral presentation

Detecting Anomalies in Space and Terrestrial Electromagnetic Data Using Sequential Data Analytics

Yaxin Bi1, Vyron Christodoulou1, Xiangzeng Kong1, Ming Huang1, Guoze Zhao2, Bing Hang2

1Ulster University, United Kingdom; 2Institute of Geology, China Earthquake Administration

In Dragon 3, we have developed a number of algorithms for detecting anomalies from sequential data and evaluated them over benchmark datasets, preliminary applications on Outgoing Longwave Radiation observed by the NOAA satellites and electromagnetic data observed by the Swarm satellites and the network of the Control Source Extremely Low Frequency (CSELF). In this report we present comprehensive and scaled up analysis results on seismic anomaly detection and relationship with the cycle of earthquakes. The report will include 1) the development of anomaly detection algorithms related to Geometric Moving Average Martingale (GMAM) method, Weighted Local Outlier Factor LOF method and Hot SAX methods, the Cumulative Sum (CUSUM) and the Exponentially Weighted Moving Average (EWMA), and two variants of a combined Cusum-EWMA; 2) comparative study methodology as well as various comparative analysis results over the Swarm and CSELF data sets.

Oral presentation

The Application of the Discrete Wavelet Method in the Electromagnetic Data in CSELF Seismological Continuous Observation Stations

Bing Han1, Guoze Zhao1, Yaxin Bi2, Yan Zhan1, Ji Tang1

1China Earthquake Administration, China, People's Republic of; 2Ulster University, UK

Two earthquakes occurred at 6th December at the vicinity of the southwestern boundary of a seismic active block (Sichuan-Yunnan block). One is Mw 5.8 earthquake at 2:43 and another is Mw5.9 earthquake at 18:20,the epicenter located at the northwest end area of Wuliangshan fault which is a left lateral strike slip fault along the northwest direction and the hypocenters depth are at 9 and 10km respectively. The CSELF station located to the northeast side of the epicenter with about 32km distance to the epicenter. In this paper we focus on analyzing the data recorded from October 18, 2014 to March 31, 2015, a total of 163 days. We tried to find some rules of electromagnetic in the change of the apparent resistivity , amplitude spectrum, raw time series and their discrete wavelet transform result respectively

  1. According to the relay of the apparent resistivity curves(get from edi files) result we find that the data in the frequency band of 0.02Hz-0.07Hz and 0.2Hz-10Hz are good and finally we choose the 0.02Hz、0.37Hz、1.2Hz、2.91Hz as typical frequencies to analysis.
  2. To analyze the varies of auto-power spectrum got from EDI files
  3. The time series of the electromagnetic field recorded in one day is save as an ats file, and the sample rate is 16Hz. We divide one day’s data into 6 periods (0-4, 4-8. 8-12, 12-16, 16-20,20-24 oclock) and calculate their averages for 6 periods and the deviations, the magnetic data are stable without strong disturbance for 0-4 oclock and electric data for 20-24 oclock and are chosen for further analyzing.
  4. We use the discrete wavelet transform method to the above stable data and select detailed scale coefficient 2, 3, 5 and 9 which are corresponding to the frequencies 0.02, 0.33, 1.33 and 2.67 Hz. Finally, the law of energy change for these frequencies can be obtained. Comprehensive Analysis of the above different parameters we can find the anomalies for these different parameters existed at almost same time period.

The project is supported by National Development and Reform Committee of China(2010-2015)and National Natural Science Foundation (41074047).

Han-The Application of the Discrete Wavelet Method in the Electromagnetic Data in CSELF Seismological_Cn_version.pdf


Symbolic Representation of Electromagnetic Data for Seismic Anomaly Detection

Vyron Christodoulou1, Yaxin Bi1, George Wilkie1, Guoze Zhao2

1Ulster University, United Kingdom; 2Institute of Geology, China Earthquake Administration

Electromagnetic data gathered by the SWARM satellite constellation is a powerful tool and gives us the opportunity to investigate their relation with seismic events under pre-defined regions of observation. In this study we examine this relation by applying different kinds of algorithms under an Anomaly Detection scope of electromagnetic sequential time series data. Two statistical based algorithms, a variant of a CUSUM-EWMA algorithm and a state of the art algorithm used by Twitter for Anomaly Detection that employs a Seasonal Hybrid Decomposition component are the first to be examined. In addition, due to the big-data nature of the study it is known that dimensionality reduction has improved results when dealing with large amounts of data. For this reason two more sophisticated methods are also analyzed. As a comparison, the widely known HOT-SAX method that uses a symbolic representation for dimensionality reduction and a new proposed and improved version of the HOT-SAX are also applied to address the sequential anomaly detection problem. This new method, uses a variable subsequence, the slope and the mean value of each subsequence to achieve the symbolic representation. The aim of this study is to investigate how symbolic representations with dimensionality reduction assist and compare to more traditional statistical methods in sequential anomaly detection. The limitations of the methods are also highlighted and some possible solutions are proposed. The algorithms have been evaluated and compared in ten different benchmark datasets. As a second evaluation step all algorithms have also been applied to real SWARM and ground-based electromagnetic sequential data. The results are presented in terms of classification by using the F-measure and the computational complexity of each algorithm.