基于GNSS监测的SSA-SVR模型边坡变形预测
Slope deformation prediction of SSA-SVR model based on GNSS monitoring
任文辉 1杨晓华 2冯永年 1杨玲 3魏静3
作者信息
- 1. 中铁建陕西高速公路有限公司,陕西西安 710018
- 2. 中国铁建投资集团有限公司,广东 珠海 519000
- 3. 北京交通大学土木建筑工程学院,北京 100044
- 折叠
摘要
针对GNSS监测数据的非平稳性和其存在的噪声会影响边坡安全变形预测的问题,以吴华高速公路超深路堑边坡为例,提出了基于平滑先验分解(SPA)和奇异值分解(SVD)消噪的麻雀搜索算法(SSA)优化支持向量机回归(SVR)的边坡变形预测模型(SPA-SVD-SSA-SVR模型),并对比分析了分解和消噪两种数据处理方式对边坡变形预测结果的影响.结果表明:该高边坡处于安全状态,整体变形较小,经SSA优化后的SVR模型(SSA-SVR模型)的预测效果较好,相较于传统SVR模型,其对监测点G1预测结果的MSE、MAE分别减小8.68％、3.82％,对监测点G2预测结果的MSE、MAE分别减小11.60％、3.26％;SPA分解和SVD消噪均可以减小GNSS监测数据的非平稳性和噪声对预测精度的影响,但单分解处理比单消噪处理的预测精度高,整合分解和消噪两种预处理的SPA-SVD-SSA-SVR模型预测效果更好,其对监测点G1预测结果的MSE、MAE分别减小31.06％、19.59％,对监测点G2预测结果的MSE、MAE分别减小28.59％、15.03％.研究结果为边坡变形监测数据的处理与边坡安全变形预测提供了新思路.
Abstract
Slope deformation prediction is an effective method to study slope stability and early warning.There exists non-stationarity of high slope GNSS monitoring data and the existing noise affects the safety analysis of the slope.We take the ultra-deep cutting slope of Wuhua Highway as a case and propose slope deformation prediction model based on sparrow search algorithm optimized for support vector regression with smooth prior analysis decomposition and singular value decomposition denoising(SPA-SVD-SSA-SVR model).The influence of two data processing methods,namely decomposition and denoising,on the predic-tion results are compared.The results show that the high slope is in a safe state with overall small deforma-tion.The SSA-SVR model demonstrates improved prediction performance.Compared to the traditional SVR model,it reduces mean squared error(MSE)and mean absolute error(MAE)by 8.68％and 3.82％,respectively,for monitoring point G1,and by 11.60％and 3.26％,respectively,for monitoring point G2.Both SPA decomposition and SVD denoising can reduce the non-stationarity and noise impact of GNSS mo-nitoring data on prediction accuracy.However,the prediction accuracy of the single decomposition process is higher than that of the single denoising process.The integrated SPA-SVD-SSA-SVR model,which com-bines decomposition and denoising,shows better prediction performance.It reduces MSE and MAE by 31.06％and 19.59％,respectively,for monitoring point G1,and by 28.59％and 15.03％,respectively,for monitoring point G2.The research results provide new insights into the processing of slope deformation monitoring data and slope safety deformation prediction.
关键词
边坡变形预测/平滑先验分解/奇异值分解/麻雀搜索算法/支持向量机回归Key words
slope deformation prediction/smoothing prior decomposition/singular value decomposition/sparrow search algorithm/support vector machine引用本文复制引用
基金项目
中国铁建投资集团有限公司科技研发项目(2020-C10)
出版年
2024
国家科技期刊平台
NSTL
万方数据