基于地层图像与时序数据的CNN-BiLSTM路堤沉降预测模型

韩同春,令狐知博,汪全震,周建,余锦地

CNN-BiLSTM prediction model for embankment settlement based on stratum images and time series data

Tongchun HAN,Zhibo LINGHU,Quanzhen WANG,Jian ZHOU,Jindi YU

表 1 融合多源信息的神经网络模型消融实验性能指标对比

Tab.1 Performance indicator comparison for ablation study of neural network model integrating multi-source information

模块 MAE RMSE R² CNN+BiLSTM （基线模型） 0.0381±0.0051 0.0571±0.0136 0.9373±0.0293 CNN+BiLSTM+ SA 0.0363±0.0068 0.0572±0.0152 0.9415±0.0172 CNN+BiLSTM+ 残差连接 0.0371±0.0039 0.0595±0.0098 0.9297±0.0266 CNN+BiLSTM+ SA+残差连接 0.0323±0.0040 0.0476±0.0084 0.9591±0.0163