Research on Coal Spontaneous Combustion Temperature Prediction Model Based on BO-LSTM

Abstract

The accurate prediction of spontaneous coal combustion temperature is crucial for early warning efforts against coal fires. Nonetheless, the challenge of enhancing prediction accuracy for these temperatures persists. This paper delves into the laws governing the variation of index gases with temperature during coal's low-temperature oxidation phase. Experimental data from temperature-programmed gas chromatography of 957 coal samples were collected, and after data cleaning, 251 valid samples date were selected. Six key indicator gases were identified through correlation heat map analysis. The ratio of the training set to the test set of the valid sample data is 67%:33%, which is applied to the models BO-LSTM, LSTM-GRU, BO-XGBoost, XGBoost, M-LSTM, and NRBO-XGBoost. The results indicate that the BO-LSTM model performs well. Additionally, the hyperparameters of the LSTM model are optimized using the Bayesian optimization algorithm to avoid falling into local optima, which significantly improves prediction accuracy.Additionally, coal samples from six mines across three regions in China were selected for validation. The results indicate that the BO-LSTM model achieves an R² value of 99.6% in predicting the spontaneous combustion temperature of coal, demonstrating its exceptional performance and potential for application.