基于小波变换和GALSSVM的边坡位移预测

边坡变形是一个受多种因素综合作用的复杂非线性动力学演化过程,用现有的物理模型来解决边坡变形的预测问题有很大难度.大量的研究工作表明,用实测的边坡位移时间序列来预测边坡未来变形更为准确,而将多种方法组合起来进行预测成为研究的主要趋势.在此基础上,建立了一种基于小波变换和进化最小二乘支持向量机(GALSSVM)的边坡位移预测模型.首先利用小波变换将边坡时间序列分解为低频分量和高频分量,然后利用互信息法和伪近邻法得到各分量的时间延迟和嵌入维数并进行相空间重构,再根据各个相空间的特点建立相应的GALSSVM预测模型,最后把各分量的预测结果进行小波重构,重构后的结果即为最终的边坡位移预测结果.对丹巴滑坡预测研究表明,这种新的预测模型具有较高的预测精度,可以应用于实际工程.

Prediction of slope displacement based on wavelet transform and genetic algorithm-least square support vector machine

The slope displacement is an explicit process of the complicated dynamic system involving many mutual facts. And the physical modeling is very difficult to fulfill prediction function. As an alternative, it is proved by many study work in which a set of displacement time series to predict the future displacement can be used; and united many different methods as a new predicting model was a principal trend of study. Based on this, a novel model based on wavelet transform and genetic algorithm -least square support vector machine (GALSSVM) for slope nonlinear displacement forcasting is proposed. Firstly, slope displacement time series are decomposed into different frequency signals through the wavelet transform. Secondly, phase space of each signals is reconstructed, and time delay and embedding dimension are determined by mutual information method and false nearest neighbor method respectively.Then, the respective forcasting model of genetic-least square support vector machine is constructed according to different characteristics of each phase space. Lastly, the predicted results of the signals are reconstructed to be used as the final prediction result of slope displacement. As a test, this model has been used in displacement prediction of DANBA slope. The results indicate that it is reliable with high precision; and it can be used to practical engineering.