Sealed-off zones in coal mines carry serious fire and explosion risk as seven gases (O₂, CO, CH₄, CO₂, H₂, N₂, C₂H₄) co-evolve. Kumari et al. (2021) propose a UMAP-LSTM monitoring and forecasting framework. Summary below.
Scenario and goals
- Signals: multichannel gas series in sealed areas.
- Goals: forecast concentrations, infer fire status and explosibility.
- Challenge: map model outputs to interpretable safety indices and diagrams.
Method highlights
1. Seven-gas joint modeling
Multivariate series capture fire evolution better than single gases, but need dimensionality reduction first.
2. UMAP + LSTM
- UMAP reduces the seven-gas series;
- LSTM forecasts future concentrations on the reduced representation.
3. Fire and explosibility
Forecast concentrations feed fire ratios (Graham, Young, CO/CO₂, C/H, JTR) and Coward / Ellicott explosibility graphics.
4. System aim
Continuous monitoring, rolling prediction, and advance warning before irreversible events.
Practical takeaways
- Prefer pipelines: reduction → sequence model → safety indices → alerts.
- Diagrams aid operator understanding beyond raw curves.
- Replicate and calibrate on each mine’s data.
Reference
Kumari, K.; et al. UMAP and LSTM Based Fire Status and Explosibility Prediction for Sealed-off Area in Underground Coal Mine. Process Safety and Environmental Protection 2021, 146, 837–852.