EPR-RCGA-based modelling of compression index and RMSE-AIC-BIC-based model selection for Chinese marine clays and their engineering application

The compression index is a key parameter in the field of soft clay engineering. In this paper, we propose an improved method for correlating the compression index with the physical properties of intact Chinese marine clays that are involved in many construction projects in coastal regions in China. First, the compression index and some common physical properties of clays from 21 regions along the Chinese coast are extracted from the literature. Then, a basic regression analysis for the compression index using the natural water content and Atterberg limits is conducted. To improve the correlation performance, an evolutionary polynomial regression (EPR) and real coded genetic algorithm (RCGA) combined technique is adopted to formulate different equations involving different numbers of variables. An optimal correlation using only natural water content and liquid limit as input parameters is finally selected according to the root mean square error (RMSE), Akaike's information criterion (AIC), and Bayesian information criterion (BIC). The proposed correlation is evaluated and shown to perform better than existing empirical correlations in predicting the compression index for all selected Chinese marine clays. This correlation is validated to be reliable and applicable to engineering applications through the prediction of the properties of an embankment on the southeast coast of China using finite element method. All comparisons show that the EPR and RCGA combined technique is powerful for correlating the compression index with the physical properties of the clay, and that model selection by RMSE, AIC, and BIC is effective. The proposed correlation could be used to update current formulations, and is applicable to engineering design in coastal regions of China.