Neural Network Model for Liquefaction Potential in Layered Soils Using Turkey and Taiwan Earthquake Data

Abstract

Simplified methods have been practiced by researchers to assess nonlinear liquefaction potential of soil. Derived from several field and laboratory tests, various simplified procedures have been developed by utilizing case studies and undisturbed soil specimens. In order to address the collective knowledge built up in conventional liquefaction engineering, an alternative general regression neural network (GRNN) model is proposed in this paper. To meet this objective, a total of 620 sets of data including twelve soil and seismic parameters are introduced into the model. The data includes the results of field tests from the two major earthquakes that took place in Turkey and Taiwan in 1999 and some of the desired input parameters are obtained from correlations existing in the literature. The proposed GRNN model was developed in four phases, mainly: identification phase, collection phase, implementation phase, and verification phase. An iterative procedure was followed to maximize the accuracy of the proposed model. The case records were divided randomly into testing, training, and validation datasets. Generating a model that takes into account of twelve soil and seismic parameters is not feasible by using simplified techniques; however, the proposed GRNN model effectively explored the complex relationship between the introduced soil and seismic input parameters and validated the liquefaction decision obtained by simplified methods. The proposed GRNN model predicted well the occurrence/nonoccurrence of soil liquefaction in these sites. The model provides a viable tool to geotechnical engineers in assessing seismic condition in sites susceptible to liquefaction.