Recurrent adversarial learning for geo-technical time-series augmentation: application to slope instability forecasting in open-pit mines

Slope instability in open-pit mines represents a critical geological engineering hazard, characterized by frequent catastrophic failures that jeopardize both operational safety and economic sustainability. Conventional data-driven displacement prediction models exhibit pronounced performance degradation under small-sample conditions, significantly impeding their practical applicability. To address this challenge, this study introduces an innovative hybrid framework integrating Recurrent Generative Adversarial Network (RGAN)-based data augmentation with Simulated Annealing (SA)-optimized Support Vector Regression (SVR). The proposed RGAN architecture synthesizes geo-technical time-series data that strictly adheres to the statistical distribution of real-world monitoring datasets, while the SA algorithm dynamically optimizes SVR hyper-parameters to bolster predictive robustness. Comprehensive experimental validation demonstrates that models trained on augmented datasets achieve a 33.16% reduction in mean absolute error (MAE) relative to baseline models employing solely original data. Sensitivity analyses further reveal an optimal synthetic-to-real data ratio of 1:1 for peak predictive performance. The principal contributions of this work are threefold: (1) development of a domain-specific RGAN architecture tailored for geo-technical time-series augmentation, (2) establishment of an integrated pipeline synergizing data generation with model optimization, and (3) provision of a scalable solution for small-sample learning in slope stability prediction. This research advances intelligent early-warning systems by proposing a data-efficient paradigm for high-risk slope monitoring applications.