Coal mine gas emission prediction based on multifactor time series method

Abstract

The prediction of coal mine gas emission is an important indicator for ventilation systems reliability and a data basis for mine gas extraction design. The traditional gas emission prediction methods have weak universal applicability, and the existing prediction models are mostly based on single-factor time series prediction. To solve this problem, the gas emission prediction method based on Recursive Feature Elimination with cross-validation (RFECV) and Bidirectional Long and Short-Term Memory (Bi-LSTM) was proposed. Aiming at the problems of numerous influencing factors, strong nonlinear characteristics and time correlation, feature selection methods based on RFECV were applied. The RFECV method embedding of two base models, Ridge Regression (Ridge) and Random Forest (RF), obtained four gas emission prediction multifactor combinations. The predictive accuracy of different models was compared with multifactor combinations when the training set accounted for 60, 70 and 80 % of the total sample. The RMSE, MAE, R2, model stability, and running time of the RF-RFECV-Bi-LSTM model were 0.2455,0.1914,0.9897,0.9431 and 12.20 s, respectively. The result indicated that the constructed prediction model had high accuracy and reliability, which can be used as a basis for the accurate prediction of gas emission in multifactor time series.