摘要
位移是边坡失稳前最显著的特征.利用历史位移时序曲线,构建了CNN-LSTM混合预测模型,引入通道剪枝技术对模型进行压缩以简化网络结构.通过对某水泥厂边坡位移监测数据进行分析预测,实验结果表明,与LSTM、CNN、剪枝后的CNN和剪枝前的CNN-LSTM预测模型相比,剪枝后的混合模型不仅预测准确率提高,而且模型泛化能力增强.
Displacement is the most significant feature before slope failure,we constructed a CNN-LSTM hybrid prediction model by using the historical displacement time series curve,and compressed the model to simplify the network structure by introducing the channel pruning technology.After analyzing and predicting the slope displacement monitoring data from a cement plant,the result shows that the hybrid model after pruning not only improves the accurate rate,but also enhances the generalization ability of the model,comparing with the prediction models of LSTM,CNN,CNN after pruning,and CNN-LSTM before pruning.
作者
郑海青
赵越磊
孙晓云
靳强
ZHENG Haiqing;ZHAO Yuelei;SUN Xiaoyun;JIN Qiang(School of Electrical and Electronic Engineering,Shijiazhuang Tiedao University,Shijiazhuang 050043,China;Hebei Jinyu Zenith Cement Co.Ltd.,Shijiazhuang 050020,China)
出处
《河南科学》
2021年第4期524-529,共6页
Henan Science
基金
国家自然科学基金项目(51674169)
河北省自然科学基金重点项目(F2019210243)
河北省高等学校科学技术研究项目(QN2019031,ZD2019140)。
关键词
边坡位移
通道剪枝
卷积神经网络
长短期记忆网络
slope displacement
channel pruning
convolutional neutral network
long short-term memory network
