Design of a Propane Ammoxidation Catalyst Using Artificial Neural Networks and Genetic Algorithms

Abstract

Artificial neural networks (ANNs) and genetic algorithms (GAs) are applied to the optimal design of a catalyst for propane ammoxidation. The mole percentages of six components of a catalyst (P, K, Cr, Mo, Al2O3/SiO2, and VSb5WSn) are used as inputs, and the activity and the acrylonitrile selectivity serve as the two outputs. This trained optimal linear combination (OLC) network is used to evaluate the yield of new catalyst compositions generated during GA optimization. The best yield of acrylonitrile found after GA optimization is 79%, which is higher than the highest yield previously reported (64%). The OLC neural network, using the acrylonitrile yield (i.e., activity times selectivity) as the output, greatly improves the simulation of the catalyst system compared to a simple, single-network architecture. In particular, whereas single-network methods can all easily reproduce the experimental patterns used for training and validation, the OLC is markedly superior for generalizing to novel catalyst patterns.