Evaluation of liquefaction susceptibility of soil using genetic programming and multivariate adaptive regression spline

Abstract

Liquefaction of soil can be considered as one of the most disastrous seismic hazards and evaluation of liquefaction susceptibility is an important aspect of geotechnical engineering. For evaluation of liquefaction potential of soil generally two variables are required, such as: (i) the seismic demand on a soil layer expressed in terms of CSR, (ii) the capacity of the soil to resist liquefaction expressed in terms of CRR. The method for evaluation of CRR is to test undisturbed soil specimens in the laboratory. The various field tests used for the liquefaction resistance of the soil are (i) Standard Penetration Test (SPT), (ii) Cone Penetration Test (CPT) , (iii) Shear Wave velocity Measurements and (iv) Becker Penetration test (BPT). Artificial intelligent techniques such as ANN, SVM, RVM are used to develop liquefaction prediction models based on in-situ database, which are found to be more efficient as compared to statistical methods. However, these techniques will not produce a comprehensive relationship between the inputs and output, and are also called as ‘black box’ system. In the present study an attempt has been made to predict the liquefaction potential of soil based post liquefaction cone penetration test (CPT) , standard penetration test (SPT) and shear wave velocity (V_s) data using multivariate adaptive regression splines (MARS) and genetic programming (GP). A comparative analysis is made among the existing methods and the proposed MARS and GP model for prediction of liquefied and non-liquefied cases in terms of percentage success rate with respect to the field manifestations. It is observed that the prediction as per MARS and GP model is more accurate towards field manifestation in comparison to other existing methods.