基于多簇回声状态网络的边坡变形预测  

Slope Deformation Prediction Based on Multi-cluster Echo State Network

作　　者：华莎 袁于思 易灿灿[2] 张磊[2] HUA Sha;YUAN Yusi;YI Cancan;ZHANG Lei(China Railway Wuhan Electrification Bureau Group Co.,Ltd.Wuhan Division,Wuhan,Hubei 430074,China;Wuhan University of Science and Technology,Wuhan,Hubei 430081,China)

机构地区：[1]中铁武汉电气化局集团第一工程有限公司,湖北武汉430074 [2]武汉科技大学,湖北武汉430081

出　　处：《水利与建筑工程学报》2022年第2期108-112,共5页Journal of Water Resources and Architectural Engineering

基　　金：国家自然科学基金(51805382);湖北省安全生产专项资金科技项目(KJZX202007003)。

摘　　要：由于边坡失稳是一个循序渐进的过程,利用时间序列分析的方法来预测边坡未来变形,有利于实现边坡的稳定性评价。相比于传统的时间序列分析方法,多簇回声状态网络(MCESN)采用动态储备池将输入信号转换为高维状态向量,选择一组最优的状态向量来表示与任务相关的输入动态。为了验证模型的有效性,采用差分整合移动平均自回归模型(ARIMA)、回归支持向量机(SVM)、长短期记忆网络(LSTM)和传统的回声状态网络(ESN)以及MCESN对三峡船闸高边坡位移进行建模与分析,通过对比均方根误差(RMSE)和复相关系数,发现MCESN的预测精度和模型泛化能力更好。结果表明,MCESN在边坡变形预测具有良好的应用前景。Because the slope instability is a gradual process,the use of time series analysis method to predict the future deformation of the slope is conducive to the realization of the stability evaluation of the slope. Compared with the traditional time series analysis method,the Multi-Cluster Echo State Network( MCESN) uses a dynamic reservoir to convert the input signal into a high-dimensional state vector,and selects a set of optimal state vectors to represent the input dynamics related to the task. In order to verify the effectiveness of the model,the article uses ARIMA,SVM,LSTM and ESN,as well as MCESN on displacement of High Slope of Three Gorges Ship Lock is modeled and analyzed,and RMSE and the Multi-correlation coefficient are compared. It is found that MCESN has better prediction accuracy and model generalization ability. The results show that MCESN has a good application prospect in slope deformation prediction.

关 键 词：边坡变形 多聚类 回声状态网络 时间序列预测 位移预测 

分 类 号：TU454[建筑科学—岩土工程]