基于时序分析及CNN-BiLSTM-AM的阶跃型滑坡位移预测
Displacement prediction of step-like landslide based on temporal analysis and CNN-BiLSTM-AM
杨进昆 1党建武 1杨景玉 1岳彪1
作者信息
- 1. 兰州交通大学电子与信息工程学院 兰州 730070
- 折叠
摘要
传统基于递归神经网络的模型对阶跃型滑坡位移预测能力不足,为解决这一问题,提出一种基于时序分析及卷积神经网络-双向长短期记忆-注意力机制(CNN-BiLSTM-AM)的滑坡位移动态预测模型.首先使用变分模态分解方法(VMD)将序列分解为趋势项、周期项和随机项.采用二次指数平滑法拟合趋势项位移,然后引入最大互信息系数法(MIC)计算各类影响因子与周期项位移相关性,对于周期项和随机项位移采用CNN-BiLSTM-AM混合模型进行多因素和单因素预测,最终累加各分量预测值得到累积位移预测结果.实验结果表明,所提方法在最终累计位移预测结果中拟合系数R2 达0.984和0.987,平均绝对误差(MAE)分别为5.334和3.947,均方根误差(RMSE)分别为6.196和4.941,相比卷积神经网络-长短期记忆(CNN-LSTM)、麻雀搜索算法-核极限学习机(SSA-KELM)和NARX方法,所提方法能够更好的捕捉监测数据的时间相关性,预测精度显著提高,可为阶跃型滑坡预警及防治工作提供参考.
Abstract
Traditional models based on recurrent neural networks have insufficient predictive capabilities for stepwise landslide displacement.To address this issue,a dynamic landslide displacement prediction model based on time series analysis and CNN-BiLSTM-AM is proposed.Firstly,the sequence is decomposed into trend components,periodic components,and random components using the variational mode decomposition(VMD)method.The trend component displacement is fitted using the second-order exponential smoothing method.Then,the maximum mutual information coefficient(MIC)method is introduced to calculate the correlation between various influencing factors and periodic component displacement.For both periodic and random component displacements,a hybrid CNN-BiLSTM-AM model is used for multi-factor and single-factor prediction.Finally,the predicted values of each component are accumulated to obtain the cumulative displacement prediction results.Experimental results show that the proposed method achieves fitting coefficients R2 of 0.984 and 0.987 in the final cumulative displacement prediction results,with average absolute errors(MAE)of 5.334 and 3.947,and root mean square errors(RMSE)of 6.196 and 4.941,respectively.Compared to CNN-LSTM,SSA-KELM,and NARX methods,the proposed method better captures the temporal correlations in monitoring data,significantly improving prediction accuracy,and providing valuable references for stepwise landslide early warning and mitigation efforts.
关键词
阶跃型滑坡/变分模态分解/注意力机制/卷积神经网络/双向长短时记忆Key words
step-like landslide/variational mode decomposition/attention mechanism/convolutional neural network/bi-directional long short-term memory引用本文复制引用
基金项目
甘肃省教育科技创新项目(221jyibgs-05)
甘肃省军民融合专项(2020JG01)
甘肃省重点研发计划(21YF5GA158)
出版年
2024
NETL
NSTL
万方数据