Information mining using artificial neural networks and “holographic research strategy”

Abstract

Holographic research strategy (HRS) and artificial neural networks (ANNs) were applied to design catalyst library for low temperature propane oxidation. In this approach, ANNs were used as black boxes to establish the relationship between the catalyst composition and the conversion. Experimental propane conversion values of a catalyst library containing catalysts with different composition were used as input data to create linearly combined neural networks. New catalyst compositions generated by HRS were virtually tested using the linearly combined neural networks. The best predicted conversion values were 54 and 89% at 100 and 150 °C, respectively. These values are higher than the highest conversions measured experimentally (37 and 84%, respectively). As emerges from results obtained HRS optimization proved to be an efficient tool in data mining, it could complement and systematize the information originated from ANNs and reveals the generalization ability of ANNs. Thus, the combined application of HRS and ANNs resulted in further information about the catalytic system investigated.