Maximum burning rate and fixed carbon burnout efficiency of power coal blends predicted with back-propagation neural network models

Abstract

Back-propagation (BP) neural network models were developed to accurately predict the maximum burning rate and fixed carbon burnout efficiency of 16 typical Chinese coals and 48 of their blends. Early stopping method was used to prevent the BP neural network from over-fitting. The generalisation performance and prediction accuracy of the neural network thus became significantly improved. Pearson correlation analysis results showed that the maximum burning rate was most relevant to coal calorific value as well as carbon and ash content. Fixed carbon burnout efficiency was most relevant to coal volatile matter, fixed carbon and calorific value. Accordingly, three-layer BP neural network models with three input factors were developed to predict the combustion characteristics of power coal blends. The BP neural network used to predict the maximum burning rate gave a relative mean error of 1.97%, which was considerably lower than that given by the quadratic polynomial regression (7.06%). Moreover, the BP neural network used to predict the fixed carbon burnout efficiency gave a relative mean error of 0.91%, which was significantly lower than that given by the quadratic polynomial regression (4.03%).