The Study of the Liquefaction Probability and Estimation of the Relative Importance of Effective Parameters Using Fuzzy Clustering and Genetic Programming

During several past decades, many researchers have conducted studies to develop the relationship between liquefaction resistance and seismic parameters of various soils. One of the methods to assess liquefaction which is widely used is the stress-based method, which was presented for the first time by Seed and Idriss (1971) and Whitman (1971). That is an experimental method and is based on experimental and local observations.
Anyway, the relationships developed by the use of the traditional method based on experimental data are limited and don’t provide a proper and stable prediction, they also dont show major weaknesses in mathematical relations, complete structure and heterogeneous space of soil and liquefaction, while soils have quite complicated structures, there is simpler methods of examination that allow unlimited development of a model for all systems, such as genetic programming (GP), artificial neural networks (ANNS) and Neuro-fuzzy inference system (ANFIS) (Shahin et al., 2003). In this study, using fuzzy clustering, it has been initially tried to divide data to groups whose members are similar and have relatively same properties. So data with similar geotechnical and seismic properties can be placed in a cluster and hence, a closer study of the data to better understand the factors affecting soil liquefaction can be possible. Next, a model was presented for time estimation of the soil liquefaction probability using the method of genetic programming. Finally, considering the results of the conducted clustering and the study of data in each cluster, the sensitivity of parameters affecting soil liquefaction and their relative importance were analyzed using nonlinear methods.