Özet:

The gross calorific value (GCV) is an essential thermal property of coal which indicates the amount of heat energy that could be released by burning a specific quantity. The primary objective of the presented study is to develop new GCV prediction models using support vector machines (SVMs) combined with feature selection algorithm. For this purpose, the feature selector RReliefF is applied to the dataset consisting of proximate and ultimate analysis variables to determine the importance of each predictor of GCV. In this way, seven different hybrid input sets (data models) were constructed. The prediction performance of models was computed by using the square of multiple correlation coefficient (R2), root mean square error (RMSE), and mean absolute percentage error (MAPE). Considering all the results obtained from this study, the predictor variables moisture (M) and ash (A) obtained from the proximate analysis and carbon (C), hydrogen (H) and sulfur (S) obtained from the ultimate analysis were found to be the most relevant variables in predicting GCV of coal, while the predictor variables volatile matter from the proximate analysis and nitrogen from the ultimate analysis did not have a positive effect on the prediction accuracy. The SVM-based model using the predictor variables M, A, C, H, and S yielded the highest R2 and the lowest RMSE and MAPE with 0.998, 0.22 MJ/kg, and 0.66%, respectively. For comparison purposes, multilayer perceptron and radial basis function network were also used to predict GCV.

Anahtar Kelimeler: