Metro Foundation Pit Deformation Prediction Method Based on CNN-GRU Neural Network under Incomplete Data Condition
[Objective] To address the issues of prediction delay and accuracy degradation caused by incomplete deformation monitoring data in metro foundation pits, a prediction method based on a CNN-GRU (convolutional neural network-gated recurrent unit) neural network is proposed and verified. [Method] A dat...
Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Article |
| Language: | zho |
| Published: |
Urban Mass Transit Magazine Press
2025-06-01
|
| Series: | Chengshi guidao jiaotong yanjiu |
| Subjects: | |
| Online Access: | https://umt1998.tongji.edu.cn/journal/paper/doi/10.16037/j.1007-869x.20230612.html |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | [Objective] To address the issues of prediction delay and accuracy degradation caused by incomplete deformation monitoring data in metro foundation pits, a prediction method based on a CNN-GRU (convolutional neural network-gated recurrent unit) neural network is proposed and verified. [Method] A data sample set is constructed using incomplete foundation pit deformation monitoring data and missing monitoring data from nearby monitoring points, and then input into the CNN model to complete data imputation and obtain continuous and complete monitoring data. Wavelet decomposition is applied to extract low-frequency trend components and high-frequency error components from the deformation data. The GRU neural network model and ARMA (autoregressive moving average) model are respectively used to predict the low-frequency trend and noise components, which are then combined to yield the final deformation prediction results. The foundation pit project at a metro station in Nanjing is used as case study to verify the effectiveness of the proposed method. [Result & Conclusion] When the proposed CNN-GRU-based prediction method is applied to foundation pit deformation data imputation with missing rates of 18.5% and 10.1%, the resulting prediction errors are 1.926 6% and 1.274 6%, respectively, and the prediction accuracy improves by 35% and 6%, respectively. These results demonstrate strong data recovery capability and high reliability of this method. Compared to the GA-BP neural network and LSTM prediction methods, the proposed method improves prediction accuracy by more than 100% and effectively addresses the issue of prediction lag. The accuracy of this method could meet the requirements of practical engineering applications. |
|---|---|
| ISSN: | 1007-869X |