Prediction of Pyrite Oxidation in a Coal Washing Waste Pile Applying Artificial Neural Networks (ANNs) and Adaptive Neuro-fuzzy Inference Systems (ANFIS)

Two models were evaluated as alternative methods for predicting pyrite oxidation in the Alborz Sharghi coal washing waste pile (in northeastern Iran). The first model applies a ‘feed-forward artificial neural network (ANN) with 4-7-1 structure’. The model uses depth, initial remaining pyrite fraction, mole fraction of oxygen, and annual precipitation as input parameters and returns the remaining pyrite fraction in the related depth of the pile as its output. In the second model, an adaptive neuro-fuzzy inference system (ANFIS), which uses generalised bell membership functions and the Takagi–Sugeno-type fuzzy inference system, was applied with the same input–output parameters. The correlation coefficient, root mean squared error, and average absolute relative error for the training stage of the ANNs were 0.81, 0.169, and 0.12, respectively, while the values for ANFIS were 0.91, 0.091, and 0.078, respectively. Comparison of the correlation coefficients and the error parameters revealed that both models successfully predicted remaining pyrite fraction from various depths of the pile. However, ANFIS was found to be more reliable and more accurate.