欢迎访问中国科学院大学学报,今天是

中国科学院大学学报 ›› 2023, Vol. 40 ›› Issue (4): 514-522.DOI: 10.7523/j.ucas.2022.002

• 地质与地球科学 • 上一篇    下一篇

基于EEMD-Prophet-LSTM的滑坡位移预测

王震豪1,2, 聂闻1, 许汉华3,4, 简文彬5   

  1. 1 中国科学院福建物质结构研究所, 福州 350002;
    2 中国科学院大学, 北京 100049;
    3 中国有色金属工业昆明勘察设计研究院有限公司, 昆明 650051;
    4云南省岩土工程与地质灾害重点实验室, 昆明 650051;
    5 福州大学岩土工程与工程地质研究所, 福州 350116
  • 收稿日期:2021-10-19 修回日期:2022-01-06 发布日期:2022-01-13
  • 通讯作者: 聂闻,E-mail:wen.nie@fjirsm.ac.cn
  • 基金资助:
    国家自然科学基金(41861134011)资助

Prediction of landslide displacement based on EEMD-Prophet-LSTM

WANG Zhenhao1,2, NIE Wen1, XU Hanhua3,4, JIAN Wenbin5   

  1. 1 Fujian Institute of Research on the Structure of Matter, Chinese Academy of Sciences, Fuzhou 350002, China;
    2 University of Chinese Academy of Sciences, Beijing 100049, China;
    3 Kunming Prospecting Design Institute of China Nonferrous Metals Industry Co., Ltd, Kunming 650051, China;
    4 Yunnan Key Laboratory of Geotechnical Engineering and Geohazards, Kunming 650051, China;
    5 Institute of Geotechnical and Geological Engineering, Fuzhou University, Fuzhou 350116, China
  • Received:2021-10-19 Revised:2022-01-06 Published:2022-01-13

摘要: 对于阶跃型滑坡位移这一非稳态自然过程,提出一种结合集合经验模态分解法(EEMD)、Prophet和长短时记忆网络(LSTM)的滑坡位移预测方法。以白水河滑坡位移数据为例,采用EEMD将位移时间序列分解为若干个本征模态函数(IMF)和残差(RES),将包含周期因素、随机因素的IMF叠加视为波动项,RES视为趋势项。分别采用Prophet和LSTM预测趋势项与波动项,两项预测结果叠加得到滑坡位移预测值。结果表明:该方法对于少量数据的白水河滑坡位移预测拟合度(R2)达到0.98以上,优于支持向量机、人工神经网络等传统机器学习方法。且此方法对八字门滑坡各监测点的预测精度R2同样在0.96以上,证明了此方法的有效性。

关键词: 滑坡位移, 时间序列, 集合经验模态分解, Prophet, 长短时记忆网络

Abstract: For the unsteady process of step-type landslide displacement, a method combining ensemble empirical mode decomposition (EEMD), Prophet, and long short time memory network (LSTM) to predict landslide displacement is proposed. The displacement data of Baishuihe landslide was taken as examples. The displacement time series was decomposed into residual(RES) and several intrinsic mode functions(IMF) by EEMD. The superimposition of IMFS which included periodic factors and random factors was considered as a volatility item, and the RES was regarded as a trend term. The trending term was fitted by the Prophet and the the volatility term was predicted by LSTM. The addition of the two prediction results was the predictied value of the landslide displacement. The results show that the coefficient of determination(R2) of the EEMD-Prophet-LSTM model is above 0.98 for Baishuihe landslide displacement prediction, which is better than traditional machine learning methods such as support vector machine and artificial neural network. Moreover, the prediction accuracy R2 of this method for each monitoring point of the Bazimen landslide is also above 0.96, which proves the applicability of this method.

Key words: landslide displacement, time series, ensemble empirical mode decomposition, Prophet, long short time memory network

中图分类号: