Shear Wave Velocity by Polynomial Neural Networks and Genetic Algorithms Based on Geotechnical Soil Properties

Shear wave velocity (V (S)) is a basic engineering soil property implemented in evaluating the soil shear modulus. In many instances it may be preferable to determine V (S) indirectly by common in-situ tests, for instance the standard penetration test. In this paper, the relation between V (S) and geotechnical soil parameters such as standard penetration test blow counts (N-160), effective stress and fines content, as well as overburden stress ratio is investigated. A new polynomial model is proposed to correlate geotechnical parameters and V (S), predicated on a total of 620 data sets, including field investigation records for the Kocaeli (Turkey, 1999) and Chi-Chi (Taiwan, 1999) earthquakes. This study addresses the question of whether group method of data handling (GMDH) type neural networks (NN) optimized using genetic algorithms could be used to (1) estimate V (S) based on specified geotechnical variables, (2) assess the influence of each variable on V (S). Results suggest that GMDH-type NN, in comparison to previous statistical relations, provides an effective means of efficiently recognizing the patterns in data and accurately predicting the shear wave velocity. The sensitivity analysis reveals that and fines content have significant influence on predicting V (S).