An ANN-based method for predicting Zhundong and other Chinese coal slagging potential

Abstract Conventional ash-component-based slagging indices and ash fusion temperatures (AFTs) successfully predict the slagging behavior of steam coals. However, these methods fail to tell the ZDc slagging potential accurately. In this paper, a back propagation artificial neural network (BP-ANN) is proposed to predict the slagging potential for both ZDc and other steam coals. Python is used for the training, verification and test of the BP-ANN. In training and verification, 12 ZDc, 2 lignite, 3 blend coal samples are analyzed and other 63 coals are collected as data. The result shows that the accuracy rate of the network based on ash components is 62.5%, close to the value of conventional slagging indices. It can only predict the slagging potential of certain coals. In contrast, modifying the network with coal properties can greatly increase the accuracy rate to 87.5%. The verification results of 14 samples are completely consistent with the actual slagging behavior, and that of 2 samples shows the slagging tendency. After that, 30 samples are collected from other papers for test. The results of 25 samples are completely consistent with the actual slagging behavior and that of 3 samples shows the slagging tendency, only 2 samples telling the wrong results.