|
Keywords
|
cFocal mechanism, Polarity, Waveform modeling, Moment tensor,
Green function, Strike, Dip, Rake, Convolutional neural network
|
|
Abstract
|
The study on faults is of particular importance in geological science. Because faults are the most important sources of earthquakes. In fact, the focal mechanism is the
geometric display of the fault slip in the earthquake, which is a plate structure with three angles of the strike, dip and rake. Calculating the focal mechanism is useful for accessing the tectonic characteristics of a region. In this study, the use of Convolutional neural network method is introduced to determine the mechanism of
earthquake. Initially, the first seismic wave got extracted to the earthquake registration stations. Ensuing the conversion of spectrogram on the earthquake signals, the entering of Convolutional neural network is applied. Finally, two models are designed for estimating three angles through the proposed method. The first model to estimate the angle of the strike and dip, the second model also has been investigated for estimating the angle of rake. The data used, which are the earthquake signals as large as four and ten kilometers. In this scientific research, the number of earthquake registration stations and their dispersion on the estimation of three angles has been studied. For this experiment, the results of accuracy of three angles on 8, 5 and 4 stations, respectively, with the azimuth distance, 45 degrees on the whole circle environment, 45 degrees on the half of the circle and 90 degrees on the whole circle
environment have been investigated. As a result, by increasing the number of stations and their uniform distribution, the accuracy of 8 stations will be better than 4 stations and also because the dispersion of 8 and 4 stations are much more uniform, the results compared to the five stations are better. These results are performed for the syntactic data and syntactic data with noise 1 and 3 db. The result of syntactic data hence, they are noise-free, are better than the syntactic data with noise. This algorithm also controls the noise well because the results are
|