A spatiotemporal correlation and attention-based model for pipeline deformation prediction in foundation pit...

Area of Science:

• Geotechnical Engineering
• Artificial Intelligence
• Civil Engineering

Background:

• Deformation prediction of adjacent pipelines in foundation pit engineering is critical for construction safety.
• Existing methods (constitutive models, grey correlation, feedforward neural networks) face challenges due to complex conditions and nonstationary, nonlinear monitoring data.
• Accurate prediction is essential to prevent structural damage and ensure project integrity.

Purpose of the Study:

• To propose a novel deep learning-based prediction model for adjacent pipeline deformation in foundation pits.
• To address the limitations of existing models in handling complex hydrological, geological, and data characteristics.
• To enhance the accuracy and reliability of deformation predictions, especially for long-term scenarios.

Main Methods:

• Formulated pipeline deformation as a multivariate time series problem.
• Developed a deep learning model integrating Convolutional Neural Networks (CNN) for spatial dependency extraction and Bi-directional Long-Short Term Memory (BiLSTM) units for temporal feature extraction.
• Incorporated an attention mechanism to dynamically adjust the weights of spatial-temporal features.

Main Results:

• The proposed model demonstrated superior performance compared to existing methods in a real-world subway project evaluation.
• Achieved significant improvements in prediction accuracy, with Adjusted R2 index increases ranging from 19.4 to 61.6.
• Showcased a substantial reduction in mean absolute error, decreasing by 51.5 to 70.3 compared to other models.
• Validated the effectiveness of capturing spatial-temporal dependencies and attention learning for improved engineering predictions.

Conclusions:

• The deep learning model effectively captures complex spatial-temporal dependencies in monitoring data.
• The attention mechanism enhances the model's ability to prioritize relevant features for accurate deformation prediction.
• The proposed approach offers a significant advancement for predicting adjacent pipeline deformation in foundation pit engineering, particularly for long-term forecasting.