Surface Deformation Monitoring and Prediction of Longtantian Open-Pit Mine Based on SBAS-InSAR and CNN-BiLSTM Techniques

The Longtantian coal mine is a nationally significant large-scale coal mine in China and is listed as one of the Yunnan Province’s coal mines for supply assurance and capacity increase, but extensive mining has caused significant surface deformation, posing safety hazards. High-precision monitoring and prediction of surface deformation are essential. This study used the small baseline subset interferometric synthetic aperture radar technique to extract surface deformation time-series data from Sentinel-1A images between 4 January 2020 and 19 January 2024. Considering the spatiotemporal heterogeneity, dynamic time warping based K-means clustering analysis was applied to divide the area into subspaces with similar deformation trends. A convolutional neural network optimized bidirectional long short-term memory (CNN-BiLSTM) prediction model was constructed using lithological properties (described by internal friction angle and cohesion), fault distance, rainfall, and deformation data. Results showed an annual deformation rate of –86.96–50.42 mm/year. The F2, F3, and F4 faults, the third section of the Longtan Formation (P<sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sub><italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">l³</i>) and the Quaternary (Q) lithological units, rainfall and historical deformation have all had an impact on the surface deformation of the mining area. The experimental results show that the CNN-BiLSTM architecture has significantly improved prediction accuracy compared to the basic LSTM model. The prediction error distribution of the original LSTM model is in the range of [–196.55, 106.91] mm, while the improved CNN-BiLSTM model reduces the error range to the range of [–58.22, 74.33] mm, with extreme error values reduced by 70.40% and 30.47%, respectively. Therefore, the research results can provide scientific data support for disaster prevention and reduction of Longtantian Coal mine, and provide reference for monitoring and forecasting of other mining areas.