A COMPARISON OF NEURAL-NETWORK PERFORMANCE FOR SEISMIC PHASE IDENTIFICATION

Traditional techniques of analysis and interpretation of seismic event s involve a series of complex steps involving sophisticated signal pro cessing as well as many manual tasks. Automating each of these steps i s an important goal of this ongoing research. The paper discusses the use of neural networks in performing phase identification, namely the discrimination of distinct seismic waves within a seismogram. The scop e is further restricted to the identification of only two of the regio nal principal phases, Pg and Lg, among the signals collected in the we stern United States. Using a database of 75 earthquakes and 75 undergr ound nuclear explosions, the performance of several types of neural ne tworks was compared. The performance of probabilistic neural network ( PNN), radial basis function (RBF) network and learning vector quantiza tion (LVQ) network is compared with a back-propagation network that co mbines the conjugate-gradient method with a weight-elimination strateg y. The results indicate that the latter outperformed all other methods tested.