ABSTRACT

Coal undergoes self-heating resulting in spontaneous combustion when exposed to oxygen in the air. The determination of various constituents within coal, especially the ultimate analysis and petrographic composition requires the use of sophisticated equipment and expertise, unlike the proximate analysis. In this study, an attempt has been made to predict the spontaneous combustion liability of Witbank coal, South Africa using both experiment and artificial neural network (ANN) based on the proximate analysis. The experimental tests show that the coal properties vary from one sample to another. The predictive models obtained from the ANN were compared with the conventional multilinear regression analysis (MLR) conducted. The obtained results from the predictive models showed that the ANN model is most suitable for the prediction of liability indices. The influence of the input parameters on the predicted liability index was investigated using a partial derivative method (PD). The PD of the moisture (M) and volatile matter (VM) are all positive indicating that an increase in M and VM will increase liability index, while the PD of the liability index with respect to ash (A) and fixed carbon (FC) are both negative indicating that as the value of A and FC decrease, the liability indices increases.

摘要

煤暴露于空气中的氧气时会发生自热，从而导致自燃。与近似分析不同，煤中各种成分的测定，尤其是最终分析和岩相成分的测定，需要使用复杂的设备和专业知识。在这项研究中，尝试使用基于近似分析的实验和人工神经网络 (ANN) 来预测南非 Witbank 煤的自燃倾向。实验测试表明，煤的特性因样品而异。从 ANN 获得的预测模型与进行的传统多元线性回归分析 (MLR) 进行了比较。从预测模型获得的结果表明，人工神经网络模型最适合于负债指数的预测。使用偏导数法 (PD) 研究了输入参数对预测负债指数的影响。水分 (M) 和挥发分 (VM) 的 PD 均为正，表明 M 和 VM 的增加将增加责任指数，而责任指数的 PD 相对于灰分 (A) 和固定碳 (FC)均为负，表明随着 A 和 FC 的值减小，负债指数增加。