Liquefaction potential analysis using hybrid multi-objective intelligence model

Abstract

Soil liquefaction is one of recognized nonlinear devastating types of ground failures associated with earthquakes. The analyses frameworks for this phenomenon have been addressed using different methods and correlated triggering factors in case histories. In the current paper, a hybrid model using imperialistic competitive metaheuristic algorithm (ICA) incorporated with multi-objective generalized feedforward neural network (MOGFFN) for the purpose of liquefaction potential analysis was assessed. The optimum hybrid ICA-MOGFFN model was applied on a diversified database of 296 compiled case histories comprising nine of the most significant effective parameters on liquefaction. The result of ICA-MOGFFN model demonstrated for 3.01%, 2.09% and 7.46% progress in the success rates for the safety factor, liquefaction occurrence and depth of liquefaction. Accordingly, the conducted precision–recall curves showed 5.08%, 1.73% and 3.92% improvement compared to MOGFFN. Further evaluations using different statistical metrics represented superior progress in performance of hybrid ICA-MOGFFN. The capability of the developed method then was approved from observed agreement with other accepted procedures. The results implied that the developed hybrid model was a flexible and accurate enough tool that can effectively be applied for the liquefaction potential analyses. Using sensitivity analyses, the most and least effective inputs on the predicted liquefaction parameters were identified.