LAUSR

Abstract Image analysis was adopted to estimate the ash content of clean coal, which contains approximately 3% ash. Seven features were identified based on the gray histogram of coal pictures. The Pearson correlation coefficients showed that most of the features had a good linear or quadratic relationship with the ash content. The polynomial regression (PR) method, polynomial regression after feature selection (PRFS), and the support vector machine optimized by particle swarm optimization (PSO-SVM) were employed for modeling. The PRFS model was selected owing to its best performance on the test data set after 10-fold cross-validation. The model was presented, and an average relative error of 4.16% was obtained for the training data set. Finally, the model was validated by the pilot run as the relative errors ranged from 0.61% to 5.57%, and the predicted ash content values matched the quick ash measurements.