Rockburst monitoring data acquired during underground coal mining are characterized by strong noise, nonlinearity, and multiscale coupling, which severely limit the predictive performance of existing models. To address these challenges, an innovative deep learning model was proposed. First, a hybrid denoising strategy combining Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) and wavelet transform was applied to enhance the quality of microseismic data. Subsequently, a trend–residual decomposition module was constructed to decouple the complex microseismic data into a trend component, representing long-term stress accumulation, and a residual component, capturing short-term fracture-induced fluctuations. On this basis, an improved adaptive multiscale noise-resilient Long Short-Term Memory (LSTM) unit was designed. A dynamic noise-control mechanism and a multiscale memory strategy were introduced to enable targeted feature extraction from the trend and residual branches, respectively. Furthermore, a multiscale interactive fusion (MSIF) module incorporating a channel attention mechanism was employed to dynamically integrate complementary information from both branches. The proposed framework was validated using field microseismic monitoring data from a northern coal mine. Experimental results demonstrated that the proposed model consistently outperformed five benchmark models across multiple evaluation metrics, achieving a recall of 88.37% with a false alarm rate of only 3.88%. These results confirm the effectiveness and robustness of the proposed approach for rockburst microseismic time-series prediction under noisy and complex conditions.

The degradation of mechanical performance in soils contaminated with crude oil has increasingly necessitated the development of effective stabilization strategies, particularly for supporting infrastructure constructed on compromised geomaterials. In this study, the influence of nano-silica on the shear strength parameters of crude oil–contaminated silty sand was systematically examined through monotonic triaxial testing. Silty sand specimens were prepared by incorporating 0%, 15%, 30%, and 40% silt into clean sand, after which each mixture was uniformly contaminated with crude oil at 8% of the dry soil weight. All contaminated specimens were stabilized using a 15% colloidal silica solution, applied at 15% of the dry soil mass, and subsequently cured for seven days to enable the formation of silica-based bonding networks within the soil matrix. The untreated oil-contaminated mixtures exhibited a marked reduction in shear strength with increasing silt content. In contrast, significant increases in shear strength were observed following stabilization with colloidal silica. The extent of improvement was found to depend strongly on the silt fraction. These findings provide new insight into nanoscale stabilization mechanisms in oil-contaminated geomaterials and highlight the potential of colloidal silica as a sustainable and effective agent for improving shear resistance in soils adversely affected by petroleum pollutants.