Research on a working face gas concentration prediction model based on LASSO-RNN time series data

Abstract

The effective prediction of gas concentration trends and timely and reasonable extraction measures can provide valuable references for gas control. The gas concentration prediction model proposed in this paper has the advantages of a large sample size and long time span for training data selection. It is suitable for more gas concentration change scenarios and can be used to adjust the data prediction length according to demand. To improve the applicability and practicability of the model, this paper proposes a prediction model based on the LASSO-RNN (least absolute shrinkage and selection operator) for mine face gas concentration based on actual gas monitoring data from a mine. First, the LASSO method is used to select the key eigenvectors that affect the gas concentration change. Second, the basic structural parameters of the RNN prediction model are preliminarily determined based on the broad strategy. Then, the MSE (mean square error) and the running time are used as the evaluation indicators to select the appropriate batch size and number of epochs. Finally, the appropriate prediction length is selected based on the optimized gas concentration prediction model. The results show that the RNN gas concentration prediction model has a better prediction effect than the LSTM (long short-term memory) prediction model. The average mean square error of the model fit can be reduced to 0.0029, and the predicted average absolute error can be reduced to 0.0084. The maximum absolute error of 0.0202, especially at the time inflection point of the change in the gas concentration curve, can better reflect the superiority of the RNN prediction model, that is, higher precision, robustness and applicability than LSTM.