Prediction Model for Indicator Gas Concentration of Coal Spontaneous Combustion in Goaf Areas based on CS-RF

Abstract

Accurate and rapid prediction of the concentration of coal spontaneous combustion indicator gases is critical for the prevention and control of coal spontaneous combustion in mines. Addressing the common issues with existing prediction models for coal spontaneous combustion indicator gas concentrations in goaf areassuch as low accuracy, slow convergence, and poor generalization abilitythis study analyzes the principles and processes of various algorithms. It proposes a hybrid prediction model that optimizes the Random Forest algorithm using the Cuckoo Search algorithm. This approach aims to overcome the randomness and limitations associated with manual parameter tuning in the Random Forest algorithm. Using data from the goaf of the 4427 working face at the Tiejishan Coal Mine as a case study, and with carbon monoxide concentration as the prediction target, the model’s applicability, superiority, and reliability were validated. The results showed that through adaptive global optimization, the CS-RF model significantly improves both prediction accuracy and efficiency. Compared to the RF model, the root-mean-square error (RMSE) and mean absolute percentage error (MAPE) were reduced by 11.85% and 23.83%, respectively. Training and prediction times were reduced by 74.83% and 31.43%, respectively, and the coefficient of determination (R2) for the linear fit increased by 11.97%; the error range relative to actual on-site monitoring values was 0.53%–6.91%. This study provides a new technical approach for the precise, rapid, and intelligent early warning of coal spontaneous combustion in coal mines.