基于分数阶傅里叶变换的黄土古滑坡变形预测分析
Prediction and analysis of deformation of loess ancient landslides based on fractional Fourier transform
寇丹晖1
作者信息
- 1. 河南省第四地质勘查院有限公司,河南郑州 450000
- 折叠
摘要
为实现滑坡变形的高精度预测,提出通过深入挖掘数据潜在相关性的变形组合预测思路来构建滑坡变形预测模型,即先采用分数阶傅里叶变换进行变形数据的分解处理,再通过改进果蝇优化算法、核极限学习机开展滑坡变形主趋势预测,并结合多尺度分析、支持向量机及变权组合等实现主趋势预测误差的补充预测.实例分析表明:分数阶傅里叶变换能有效实现滑坡变形数据的分解处理,且通过阶次p的优化筛选,当其值为0.3时分解效果相对较优;在变形数据分解基础上,通过组合预测得到4个监测点的相对误差平均值范围为2.01％～2.15％,训练时间范围为164.08～173.68 ms,表明提出的预测思路不仅具有较高的预测精度、较快的训练速度,还具有较强的预测稳定性,验证了预测思路的合理性,且4个监测点的速率均值范围为1.74～2.26 mm/期,增加速率也较大,无收敛特征,因此应尽快开展滑坡防治处理.研究可为类似滑坡变形预测提供借鉴.
Abstract
In order to achieve high-precision prediction of landslide deformation,a deformation combination prediction approach is proposed to construct a landslide deformation prediction model by deeply exploring the potential correlation of data.Firstly,the fractional Fourier transform is used to decompose the deform-ation data,and then the improved fruit fly optimization algorithm and kernel limit learning machine are used to predict the main trend of landslide deformation,combined with multi-scale analysis support vector machines and variable weight combinations are used to achieve supplementary prediction of main trend pre-diction errors.The analysis shows that fractional Fourier transform can effectively decompose landslide de-formation data,and through optimization and screening of order p,the decomposition effect is relatively better when its value is 0.3.On the basis of deformation data decomposition,the average relative error range of the four monitoring points was obtained through combined prediction,ranging from 2.01％to 2.15％,and the training time range was 164.08 ms to 173.68 ms.This indicates that the prediction method pro-posed in this article not only has high prediction accuracy and fast training speed,but also has strong pre-diction stability.This verifies the rationality of the prediction method proposed in this article,and the aver-age rate range of the four monitoring points is 1.74 mm/period to 2.26 mm/period,with a significant in-crease rate.There is no convergence characteristic,and landslide prevention and treatment should be carried out as soon as possible.Through this study,experience has been accumulated for predicting deformation of similar landslides.
关键词
黄土滑坡/极限学习机/数据分解/变形预测/组合预测Key words
Loess landslide/Extreme learning machine/Data decomposition/Deformation prediction/Com-bination prediction引用本文复制引用
出版年
2024
NETL
NSTL
万方数据