Innovative prediction model of carbon monoxide emission from deep mined coal oxidation

Due to great impacts to air pollution caused by residual coal oxidation in underground mine gob, monitoring and forecasting of hazardous gases emissions have become important topics in mining engineering and environmental research today. This paper presents a robot monitoring system for carbon monoxide emission from coal oxidation in spontaneous combustion condition. According to the terahertz-wave absorption spectrum, the CO concentrations are measured by using terahertz time-domain spectroscopy (THz-TDS) technique. Based on the measured values, an innovative method of CO concentration prediction has been developed by using least square support vector machine (LSSVM) with a novel hyper-parameter selection. The hourly CO concentrations have been predicted using the SVM and the hybrid LSSVM models respectively. Results show that the hybrid LSSVM has better accuracy. Statistic estimators have been employed to compare performances of the models. It has been concluded that the errors decrease and coefficients of determination increase for hybrid LSSVM model, hence it has definite practice significance and application value.