Prediction of direct coal liquefaction residue catalytic gasification based on back propagation neural network

Abstract

ABSTRACT As an important clean coal technology, direct coal liquefaction generates a large amount of residue while producing oil. Catalytic gasification is an attractive conversion route for direct coal liquefaction residue (DCLR). Due to the complex reaction process affected by many variables, so this paper applies BP neural network which has the advantage of solving nonlinear problems. Firstly, a back propagation (BP) neural network model was established with four inputs (the relative molecular weight of catalyst, the melting point of catalyst, the loading content of catalyst, and the final gasification temperature) and three outputs (the gasification conversion, the maximum gasification rate, and gasification reaction index reaction) to simulate the catalytic gasification of DCLR in CO2 atmosphere. In order to enhance the training speed and the stability of the network, some improvement of BP model were conducted by both additional momentum term and adaptive learning rate, or variable excitation function, and the prediction relative errors of the two models are less than 5%. Finally, the impact degree of factors was investigated by weight analysis based on the BP single-output model, and the result shows that the catalyst type is the most significant among these variables.