基于CEEMDAN的深度学习滑坡位移组合预测模型  

A combined landslide displacement prediction model based on CEEMDAN and deep learning

作　　者：舒玉平 徐金浩 SHU Yuping;XU Jinhao(Zhejiang Academy of Surveying and Mapping,Hangzhou,Zhejiang 311100,China)

机构地区：[1]浙江省测绘科学技术研究院,浙江杭州311100

出　　处：《北京测绘》2025年第4期554-560,共7页Beijing Surveying and Mapping

基　　金：浙江省2023年度自然资源科技项目(2023-33)。

摘　　要：本文构建了一个融合时间序列分解与分量重组的滑坡深度学习组合预测模型。首先,利用孤立森林算法剔除监测数据中的粗差点,分析数据的平稳性、自相关性以及正态分布特征;其次,引入自适应噪声完备集合经验模态分解方法(CEEMDAN),将滑坡监测数据精细分解为多个独立的时序分量;最后,针对不同频段分量制定预测模型,重构预测结果。基于北斗卫星系统(BDS)/全球导航卫星系统(GNSS)系统采集的滑坡体数据样本进行实验,结果表明,较单一模型、分解模型,本文提出组合预测模型的R^(2)值分别提升了60.66%、50.77%,均方根误差(RMSE)S_(RMSE)分别降低了95.42%、94.39%,平均绝对误差(MAE)S_(MAE)分别降低了95.69%、96.74%。This paper presented a combined landslide deep learning prediction model that integrated time series decomposition and component reconstruction.First,the isolated forest algorithm was used to remove outliers from the monitoring data,and the data's stationarity,autocorrelation,and normality were analyzed.Next,the adaptive noise complete ensemble empirical mode decomposition with adaptive noise(CEEMDAN) method was introduced to decompose the landslide monitoring data into multiple independent time series components.Finally,prediction models are developed for different frequency components,and the prediction results are reconstructed.Experiments were conducted using landslide data samples collected from the Beidou satellite system(BDS)/global navigation satellite system(GNSS).The results show that the R~2 value of the proposed combined prediction model is improved by 60.66% and 50.77% compared to the single model and the decomposition model,respectively.The mean-root-square error S_(RMSE) is reduced by 95.42% and 94.39%,and the mean absolute error S_(RMSE) is reduced by 95.69% and 96.74%,respectively.

关 键 词：自适应噪声完备集合经验模态分解(CEEMDAN) 样本熵 深度学习 滑坡位移预测 

分 类 号：P258[天文地球—测绘科学与技术]