Abstract: In view of the nonlinearity and complexity of deformation series of foundation pit, we propose to research the deformation law of foundation pit by using multifractional detrended fluctuation analysis (MF-DFA) and extreme learning machine improved by particle swarm optimization (PSO-ELM). First of all, we adopt MF-DFA method to analyze the series of deformation rate of foundation pit; secondly, we employ PSO-ELM model to process the cumulative deformation series of foundation pit; finally, we can determine the comprehensive deformation trend of foundation pit by comparing the results of both deformation series. Conclusions imply that MF-DFA could effectively reflect the multifractional feature of deformation rate series, and meanwhile PSO-ELM model is of high accuracy in predicting deformation. The analysis results of the two methods are well consistent, which supports each other in accuracy.

Key words: foundation pit, MF-DFA, extreme learning machine, deformation trend judgment, deformation prediction