A dynamic model for the seismic signals processing and application in seismic prediction and discrimination

In this paper we have presented a new method for seismic signal analysis, b ased on the ARMA modeling and a fuzzy LVQ clustering method. The objective achieved in this work is to sense the changes made naturally or artificiall y on the seismogram signal, and to detect the sources, which caused these c hanges (seismic classification). During the study, we have also found out t hat the model is sometimes capable to alarm the further seismic events just a little time before the onset of those events (seismic prediction). So th e application of the proposed method both in seismic classification and sei smic prediction are studied through the experimental results. The study is based on the background noise of the teleseismic short period recordings. T he ARMA model coefficients are derived for the consecutive overlapped windo ws. A base model is then generated by clustering the calculated model param eters, using the fuzzy LVQ method proposed by Nassery & Faez in [19]. The t ime windows, which do not take part in model generation process, are named as the test windows. The model coefficients of the test windows are then co mpared to the base model coefficients through some pre-defined composition rules. The result of this comparison is a normalized value generated as a m easure of similarity; The set of the consecutive similarity measures genera te above, produce a curve versus the time windows indices called as the cha racteristic curves. The numerical results have shown that the characteristi c curves often contain much vital seismological information and can be used for source classification and prediction purposes.