Analysis and testing of high-frequency regional seismic discriminants. Final report, 27 August 1991-31 July 1992

Technical Report · Wed Sep 30 04:00:00 EDT 1992

OSTI ID:6848814

Pulli, J J; Dysart, P S

The objective of this research was to test the applicability of artificial neural networks (ANNs) for seismic event identification. Research has focused on 101 small events within regional distances of the NORESS array for which independent source identifications are available. Twelve signal parameters were extracted from the P-, S-, and Lg-waves of each event. A number of descriptive statistics were calculated for the purpose of understanding the parameter dataspace. These included means, variances, tests for normality, and cross-correlations between signal parameters. A Stepwise Discriminant Analysis was performed to eliminate parameters which do not contribute to the identification. The six most important parameters were the Pn/Lg spectral ratio from 5-10 Hz, the mean cepstral variance, the Pn/Sn wideband spectral ratio, the Pn/Lg spectral ratio from 2-5 Hz, the Pn/Sn spectral ratio from 2-5 Hz, and the Pn/Lg spectral ratio from 10-20 Hz. Principal Components Analysis was applied to the reduced parameter dataset to aid in the design of the ANN. Three principal components account for more than 85% of the data variance, providing a useful guide to the appropriate hidden layer dimensionality. The final ANN design consisted of an input layer with 12 units, a single hidden layer with three units, and an output layer of one unit. The ANN was trained using the backpropagation learning algorithm. The ANN correctly identified all but two of the 101 events, two explosions, were misclassified as earthquakes. Experiments were also conducted using inputs from 1, 5, 10, 15, 20, and 25 NORESS elements as a means of testing the ANN sensitivity to SNR. Identification performance increased from 77% for one element to 98% for 25 elements.

• Research Organization: Radix Systems, Inc., Rockville, MD (United States)
• OSTI ID: 6848814
• Report Number(s): AD-A-257763/3/XAB; CNN: F19628-91-C-0111
• Country of Publication: United States
• Language: English

Related Subjects

45 MILITARY TECHNOLOGY, WEAPONRY, AND NATIONAL DEFENSE
450300* -- Military Technology
Weaponry
& National Defense-- Nuclear Explosion Detection
58 GEOSCIENCES
580000 -- Geosciences
ALGORITHMS
DATA
DESIGN
DETECTION
EARTHQUAKES
ECONOMIC ANALYSIS
ECONOMICS
EXPERT SYSTEMS
EXPLOSIONS
INFORMATION
INPUT-OUTPUT ANALYSIS
MATHEMATICAL LOGIC
MEASURING INSTRUMENTS
NEURAL NETWORKS
NUCLEAR EXPLOSION DETECTION
NUCLEAR EXPLOSIONS
NUMERICAL DATA
P WAVES
PARTIAL WAVES
REGIONAL ANALYSIS
S WAVES
SEISMIC ARRAYS
SEISMIC DETECTION
SEISMIC EVENTS
SEISMIC WAVES
SEISMOLOGY
SENSITIVITY
SIGNALS
STATISTICAL DATA
UNDERGROUND EXPLOSIONS