USING STACKED GENERALIZATION ENSEMBLE METHOD TO ESTIMATE SHEAR WAVE VELOCITY BASED ON DOWNHOLE SEISMIC DATA: A CASE STUDY OF SARAB-E-ZAHAB, IRAN

摘要

The proper estimation of shear wave velocity (Vs), because of its direct relation to the soil dynamic properties, for the study of Site effects is an important task in the engineering geophysics. Because of the direct travel of waves from sources to receivers, the downhole seismic method, among others, is suitable for accurate estimation of shear wave velocity. However, the main challenge is the high cost of borehole surveys, which limits the amount of downhole seismic data when studying a large area. In order to tackle this problem, an ensemble system is proposed that estimates the shear wave velocity using a limited amount of data. For this purpose, the downhole seismic data at 4 points were collected in Sarab-e-Zahab area, Iran. Then, the data were processed and the shear wave velocity profile was obtained for each borehole. Finally, using an ensemble of neural networks, a 3- and 2-dimensional model of Vs was constructed for the study area. Feed-forward neural networks were used as the base classifiers in an ensemble system and two methods, namely averaging and stacked generalization were employed to combine the results of base classifiers. The performances of the two methods were compared and the shear wave velocity was estimated as a function of depth. The results of the ensemble neural networks method in the study area were compared with Kriging geostatistical method. The results show the ensemble neural networks in the Vs modeling in comparison to the Kriging method has better performance. Also, the findings showed that the stacked generalization method outperformed the averaging method in the estimation of shear wave velocity.