Prediction of flammability limit temperatures from molecular structures using a neural network-particle swarm algorithm

A hybrid method is proposed to predict thermal properties (lower flammability limit temperature and upper flammability limit temperature) of organic compounds. The model includes a simple group contribution method implemented in an artificial neural network that replaces standard back-propagation with particle swarm optimization. Several architectures were studied, and the optimum was determined. The capabilities of this neural network were tested using 328 compounds in the correlated set and 90 compounds in the predicted set. The results show that the proposed method can estimate the flammability limit temperatures of organic compounds with low deviations and can be used with acceptable accuracy in thermal engineering.