Abstract:Landslides are characterized with complex nonlinear-dynamic behavior involving many uncertain factors. The physical-based modeling approach is often very difficult to fulfill. As an alternative,based on the time series analysis theory and the idea that the displacement is one of the most important information reflecting the sliding state during the evolution of landslides,a new hybrid evolutionary method,combining genetic algorithm and genetic programming,was proposed to identify the evolution character of landslides from the observed displacement time series. In this method,the model structure and model parameters are evolved by using the symbol regression techniques of genetic programming and genetic algorithm,respectively,and a global optimal nonlinear dynamic input-output model for predicting the state of landslides is fulfilled through data analysis. Models of input and output are the displacements history and future displacements,respectively. Applications to the evolution analysis of the Xintan landslide and Bachimen landslide were performed and the results proved the efficiency of the new method. Furthermore,the new algorithm shows significant power of self-organization.