关键词: 岩土工程, 时间序列, 基础沉降, 变形预测, 遗传算法, 遗传规划, 模型识别

Abstract: The settlement behavior of constructions can be characterized as nonlinear dynamic evolution progress. Based on the theory of time series analysis and combining it with genetic evolutionary algorithms for global optimization, a new hybrid method is proposed to identify the dynamic evolution prediction model of building settlement due to the deformation data observed. According to the extremely complex search progress of model structure coupling with relevant parameters, this method thus divides the whole search progress into two relatively simple processes, for modeling, i.e., the structure evolution and parameters evolution, which are to be implemented by the symbol regression techniques of genetic programming and genetic algorithm, respectively. Then, an overall evolution scheme including both as above is designed to search the globally optimal nonlinear dynamic input/output model for predicting the building settlement. Applications to the settlement prediction of several high buildings showed that the method proposed is available to accurate prediction with extendible prediction, especially the new algorithm presents a high self-organizing ability during applications so as to provide an efficient analysis means for the prediction of settlement of high buildings.