Application of a genetic algorithm and a neural network for the discovery and optimization of new solid catalytic materials

In the process of discovering new catalytic compositions by combinatorial methods in heterogeneous catalysis usually various potential catalytic compounds have to be prepared and tested. To decrease the number of necessary experiments an optimization algorithm based on a genetic algorithm for deriving subsequent generations from the performance of the members of the preceding generation is described. This procedure is supplemented by using an artificial neural network for establishing relationships between catalyst compositions-or more general speaking-materials properties and their catalytic performance. By combining a trained neural network with the genetic algorithm software virtually computer experiments were done aiming at adjusting the control parameters of the optimization algorithm to the special requirement of catalyst development. The approach is illustrated by the search for new catalytic compositions for the oxidative dehydrogenation of propane. (C) 2003 Elsevier B.V. All rights reserved.