LSTM based encoder-decoder for short-term predictions of gas concentration using multi-sensor fusion

Abstract

Gas is one of the most dangerous byproducts of coal in mines. Before gas accidents occur, an abnormally increased gas concentration can be observed. Therefore, a prediction of the gas concentration in coal mines is of great significance to prevent the gas accident and ensure the production safety in the mines. By calculating the Pearson correlation coefficient for the gas concentration of different sensors, the spatial correlation of the gas concentration that is monitored for each mining face is verified. We present multi-step prediction results for gas concentration time series based on the ARMA model, the CHAOS model and the Encoder-Decoder model (single-sensor and multi-sensor) and compare these results. The Encoder-Decoder model provides high robustness in a multi-step prediction and can predict the gas concentration for five different time steps. Its prediction error is significantly lower than those of the ARMA and the CHAOS models. The prediction accuracy is further improved through a fusion with information of other sensors. In this way, this study provides a novel concept and method for gas accident prevention.