Artificial neural network-aided design of a multi-component catalyst for methane oxidative coupling

Abstract

Based on the properties of neural network, an improved back-propagation network, including the structural organization, the training method and the generalization ability of network, was developed to simulate the relations between components of catalyst and aspects of catalytic performance, which include C 2 selectivity and conversion of methane. Levenberg–Marquardt method is presented to train the network and get better results than are available with traditional gradient method. The catalyst, which was found by SWIFT method, was proved to be better than any other catalyst in training pattern. When reacting on the optimum catalyst, GHSV was 33,313 cm 3 g −1 h −1, CH 4:O 2 was 3, reaction temperature was 1069 K, CH 4 conversion was 27.54%, C 2 selectivity reached 75.40% (C 2 yield was 20.77%) and the activity of catalyst did not decrease obviously in 10 h.