Uncertainty analysis of liquefaction-induced lateral spreading using fuzzy variables and genetic algorithm

One of the most destructive phenomena occurring during earthquakes is liquefaction-induced lateral spreading. So far, various prediction models have been proposed focusing only on the accuracy of the lateral spreading estimations. However, the reliability of predictive models is also essential since the uncertainties of input parameters result in uncertain estimates. To assess the performance of such models under uncertainty, the values of governing parameters from a well-known database are first fuzzified. Then, the extreme amounts of the liquified soil displacement are computed via coupling the genetic algorithm with the prediction models. This is accomplished through solving optimization problems of many objectives for analysis of the uncertainty effects by the fuzzy sets theory. Considering the fuzzy statistical indices, it is found that the slightly uncertain inputs can drastically influence the responses. Moreover, since the recent model tree (MT)-based predictive model is revealed to be vulnerable against uncertainty, a novel model named MT (New model) was developed. It is concluded that both the accuracy and reliability of the proposed relationships compare favorably with other available models.

Automated summary

When predicting liquefaction-induced lateral spreading during earthquakes, the uncertainties of input parameters should be considered for the accuracy and reliability of the results. It is found that even slightly uncertain inputs can significantly influence the responses. The newly developed MT (New model) shows more reliable performance under uncertainty, with favorable accuracy and reliability compared to other models.