基于多特征融合的GraphHeat-ChebNet隧道形变预测模型

对隧道的形变进行预测是隧道结构异常检测的内容之一。为了充分挖掘形变特征的时空关联性，针对隧道内衬多个断面的形变同时预测，提出一种基于多特征融合的GraphHeat-ChebNet隧道形变预测模型。所提模型中利用GraphHeat和ChebNet这2种图卷积网络(graph convolution net,GCN)分别提取特征信号的低频和高频部分，并获取形变特征的空间关联性，ConvGRUs网络用于提取特征在时间上的关联性，通过三阶段融合方法保留挖掘的信息。为了解决实验数据在时间维度上不充足的问题，引入双层滑动窗口机制。此外，所提模型与其他模型或算法在不同数据集上实验比较，衡量一天和两天预测值的误差指标优于其他模型，而且对大部分节点预测的误差较低。说明模型受样本节点数影响较小，能较好地预测一天和两天的形变，模型学习特征与时空模式的能力较强，泛化性较好。

Tunnel deformation prediction model based on multi-feature fusion and GraphHeat-ChebNet

Prediction of tunnel deformation is one of the contents of tunnel structure abnormality detection, and the prediction of this paper is deformation displacement. In order to fully explore the temporal and spatial correlation of deformation features and simultaneously predict the deformation of multiple sections of the tunnel lining, this paper proposes a GraphHeat-ChebNet tunnel deformation prediction model based on multi-feature fusion. In the proposed model, GraphHeat and ChebNet extract the low-frequency and high-frequency parts of the deformation feature signals respectively, and obtain the spatial correlation of the features. The ConvGRUs network is used to extract the temporal correlation of the features, and the three-stage fusion is used to retain the mined information. At the same time, in order to solve the problem of insufficient experimental data in the time dimension, a double-layer sliding window mechanism is introduced. In addition, compared with other models or algorithms on different data sets, the error index of measuring the one-day and two-day prediction values is better than other models, and the error of prediction for most nodes is lower. It shows that the model is less affected by the number of sample nodes, can better predict one-day and two-day deformation, has strong ability to learn characteristics and spatio-temporal patterns, and has good generalization.