Multi-Source Geological Data Fusion and Deep Learning-based Displacement Forecasting for Sustainable Geotechnical Monitoring

This study addresses the challenges of geological parameter inversion and displacement prediction in geological engineering by proposing a deep learning algorithm based on multi-source geological data fusion. This algorithm constructs a fusion framework encompassing multi-modal, multispatiotemporal scale, and multi-source heterogeneous data. At the theoretical level, it combines a convolutional neural networklong short-term memory network (GA-CNN-LSTM) optimized by a genetic algorithm with Physics Informed Neural Networks (PINN) constraints. This optimization achieves high-precision parameter inversion and displacement trend prediction in complex geological environments. Specifically, the research system integrates multidimensional data such as GNSS deformation, fiber-optic distributed strain, meteorological data, groundwater seepage, and remote sensing topography. The model constructs a unified high-dimensional feature space through multi-level spatiotemporal alignment and weighted fusion. Based on this, the GA algorithm automatically optimizes the deep network hyper-parameters to enhance model performance. Furthermore, the PINN introduces physical constraints from elasticity and seepage governing equations. This fusion effectively improves the physical plausibility of the predictions. Experiments using self-collected time-series data from a landslide monitoring site in Hebei Province demonstrate that this method achieves a correlation coefficient of 0.94 for geological parameter inversion, significantly outperforming traditional physical methods and single-data source models. At the same time, in the 7 -day displacement prediction task, the PINN-integrated model achieved a root mean square error (RMSE) of 5.1 cm, a maximum deviation of 9.8 cm, and a prediction stability of 2.7 cm. The method proposed in this study fully verifies the application potential of multi-source data fusion and deep physical constraints in geological engineering safety monitoring.