Comparative analysis and research on the prediction performance of gas desorption volume based on five machine learning algorithm models

The control of gas is a crucial aspect of ensuring safe production in coal mines. Accurately predicting the gas desorption volume can help detect the risk of gas outbursts in advance and guarantee safe production in coal mines. This paper selects the type of coal mine, coal sample mesh size, temperature, axial pressure, and gas pressure as input variables, and gas desorption volume as the output variable, based on the experimental results of previous studies. The study compares and analyzes the predictive performance of random forest, extreme gradient boosting (XGBoost), long short-term memory (LSTM), gated recurrent unit (GRU), and multilayer perceptron on gas desorption volume. The results show that there are certain differences in the prediction performance of different prediction models. Among them, the XGBoost model performed the best, with a determination coefficient (R2) of 0.9972. In contrast, the GRU model is relatively unstable, with the lowest R2 value of 0.884 38. In the test set, the LSTM prediction model achieved the best performance, with a mean absolute error of 0.582 18, a mean absolute percentage error of 0.191 68, a mean square error of 0.485 01, root mean square error of 0.696 43, and R2 of 0.849 08. The Shapley additive interpretation method revealed that gas pressure has the greatest influence on the gas desorption volume. This research can mitigate data interference to a certain extent for the application of machine learning. It also provides some significant guidance for controlling coal mine gas and ensuring safe production.