Intelligent prediction of minimum spouting velocity of spouted bed by back propagation neural network

A back-propagation neural network (BP-neural network) model was developed to predict the minimum spouting velocity (Ums) in spouted beds. Five dimensionless variables involving seven key geometric and operating parameters of spouted beds, i.e. column diameter, spout nozzle diameter, base angle, static bed height, particle diameter, particle density and gas density, were constructed as model inputs. An adaptive genetic algorithm was used to determine the nuclear parameters in a BP-neural network. 164 experimental data from the published literature were divided into two equal groups, for training and validating the neural network model, respectively. Comparisons of predictions by the BP-neural network and existing empirical equations with experimental data showed that Ums values predicted by the BP-neural network were in good agreement with experimental values, with a mean relative error of 12.9%, somewhat better than calculations by existing empirical equations. This indicates that an artificial neural network is a useful and promising way to predict Ums as an alternative to empirical equations, especially when the relationship of geometric and operating parameters to Ums is complex and difficult to describe.