The deep learning model prediction of mining fugitive methane (VAM) driven by carbon peak and carbon neutrality

Ventilation air methane (VAM) contributes more than 80 % of the total direct emissions of coal mine methane (CMM) in China and plays the main role in the mine fugitive gas, with a single mine shaft emitting VAM at a rate of 222 m³ /s (or 800,000 m³/h) potentially generating around 1 million tons of CO₂-equivalent (CO₂e) annually. It is pivotal to quantify the VAM and strategically mitigate the emission towards carbon peak and neutrality. In this study, the Long-Short-Term-Memory (LSTM) recurrent neural network is optimized based on the inspiration generated by the chromosomal inheritance pattern of human genes. Multiple factors with a strong correlation of CH₄ emissions are predicted, and future CH₄ emissions are predicted using time series and multi-factor series. Combined with the self-determined pattern of VAM emissions, a correlation model is proposed for VAM quantification. Based on the predicted VAM emission, recommendations are summarized for effective VAM mitigation. This work will lay the foundation for the future fugitive gas emissions from the coal mining sector for the coal-producing countries.