Prediction model of coal mine face gas concentration interval based on GRU and non-parametric kernel density estimation

Accurate gas concentration prediction is essential for preventing and controlling gas-related disasters in coal mines.However, existing methods often lack accuracy, generalizability, and rely solely on point predictions.To address these challenges, we propose an interval prediction model that combines a gated recurrent unit (GRU) network with nonparametric kernel density estimation (NKDE).The GRU network first generates point predictions of gas concentration, from which we create a dataset of prediction errors.We then use kernel density estimation, optimizing the bandwidth based on mean squared error, to model the distribution of these errors.By integrating the GRU predictions with the estimated error distribution, we construct prediction intervals at various confidence levels.We validate our approach using historical data from two different coal mine sites.Results show that the GRU model outperforms support vector machines (SVM), backpropagation neural networks (BP), and ARIMA models in prediction accuracy.Additionally, the optimized kernel density estimation provides a more accurate representation of prediction errors than both the Gaussian model and kernel density estimates with arbitrary bandwidths, resulting in more reliable prediction intervals.This method offers a practical tool to enhance safety management in coal mines.