An early warning method for high-energy mine seismic events based on spatiotemporal clustering and fractal dimension analysis

In deep mining environments characterized by elevated in-situ stress, high-energy seismic events (HE events) are closely associated with rockbursts and roof falls, posing significant threats to personnel safety. Conventional early warning methods are primarily based on static time window analyses, which inadequately capture the rapidly evolving nature of microseismic activity and often suffer from high false alarm rates. A novel early warning approach for HE events based on a dynamic analytical framework integrates kernel density based spatiotemporal clustering with fractal dimension analysis. The time windowing mechanism adopts a 5 day historical window combined with an 8 h real-time resolution, optimized through semivariogram analysis, enabling fine-grained tracking of seismic evolution. A co-evolutionary pattern is observed whereby increases in the kernel density peak are consistently accompanied by decreases in fractal dimension D, providing dual precursory indicators. A joint early warning model based on a sliding window structure is established through grid search optimization, which effectively reduces the false alarm rate compared to single-indicator models. When applied to microseismic data from the 6306 working face of the Dongtan Coal Mine, the proposed method successfully predicted 14 out of 17 HE events, achieving a recall of 0.824, a precision of 0.538, and an F1 score of 0.651. The dynamic early warning framework represents a significant step toward intelligent and adaptive early warning systems and demonstrates strong potential for pratical implementation and scalability across different mining sites.