Prediction model with multi-point relationship fusion via graph convolutional network: A case study on mining-induced...

Area of Science:

• Geosciences
• Mining Engineering
• Remote Sensing

Background:

• Accurate prediction of mining-induced surface subsidence is crucial for infrastructure safety and environmental management.
• Existing models often overlook the spatial correlations between subsidence points, leading to prediction inaccuracies.
• The study addresses the need for more robust subsidence prediction models that account for inter-point relationships.

Purpose of the Study:

• To propose and evaluate a novel Multi-point Relationship Fusion Graph Convolutional Network (MRF-GCN) model for predicting mining-induced surface subsidence.
• To improve the accuracy of surface subsidence prediction by incorporating the correlation between observation points.
• To demonstrate the model's applicability in real-world mining scenarios.

Main Methods:

• Utilized InSAR (Interferometric Synthetic Aperture Radar) data from Sentinel-1A and GNSS (Global Navigation Satellite System) observations for surface deformation analysis.
• Employed a Long-Short Term Memory (LSTM) encoder to capture deformation patterns of individual points.
• Developed an MRF-GCN model that integrates point correlations into a graph structure for enhanced prediction.

Main Results:

• The MRF-GCN model achieved a high coefficient of determination (R²) of 0.8650.
• The model demonstrated a significantly lower Mean Squared Error (MSE) of 1.59899 compared to conventional models.
• MRF-GCN exhibited superior prediction accuracy over standard Long-Short Term Memory (LSTM) and other traditional methods.

Conclusions:

• The MRF-GCN model offers a significant advancement in predicting mining-induced surface subsidence.
• The model's ability to fuse multi-point relationships enhances prediction accuracy for large-scale areas.
• This approach provides a reliable tool for analyzing and mitigating risks associated with mining-induced surface deformation.