Modeling of gross calorific value based on coal properties by support vector regression method

Gross calorific value (GCV) is one the most important coal combustion parameters for power plants. Modeling of GCV based on coal properties could be a key for estimating the amount of coal consumption in the combustion system of various plants. In this study, support vector regression (SVR) as a powerful prediction method has been used to investigate relationships among coal sample properties with their GCVs for a wide range of records. Variable importance measurement by the SVR method throughout various coal analyses (proximate, ultimate, different sulfur types, and petrography) indicated that carbon, ash, moisture, and hydrogen contents are the most effective variables for the GCV prediction. Two models based on all variables and four the most effective ones are conducted. Outputs in the testing stage of both models verified that SVR can predict GCV quite satisfactorily where the correlations of determination (R²) for models was 0.99. Based on these results, development of a variable selection system among wide range of parameters, and also application of an accurate predictive model such as SVR, can potentially be further employed as a reliable tool for evaluation of complex relationships in earth and energy problems.