Improving Continuous Coal Pyrolysis by Neural Network Control

Abstract

The use of neural networks to improve the continuous pyrolysis of coal is discussed. In particular, attention focuses on the use of a two-layer fully connected neural network to control the process parameters and boost the efficiency of coal processing. The neural network is trained by using a backpropagation algorithm on the basis of experimental data, including data on the temperature, pressure, porosity, and product yield. The neural network permits more efficient fuel processing, reduces harmful emissions, and improves product quality. The network architecture and its training by means of real time data are described. The experiments indicate that coal pyrolysis consists of two competing processes: the destruction of the coal’s organic mass; and the condensation of carbon from the gas phase on the coke that forms. The results show that artificial intelligence has great potential for improving coal processing and creating more efficient and environmentally benign processing methods.