A neural network topology for modelling grain drying

This paper is concerned with modelling moisture distribution in agricultural fixed-bed dryers using a neural network (NN). Ten different NN topologies were studied for modelling and the most appropriate one was selected to use. Inlet and outlet air temperatures, absolute humidities and air flow were considered as the input variables to the layers of the drying bed. Some topologies include also grain temperature for better performance. Randomly varying time series data simulating inlet conditions were used for training the neural network. The data were taken from a physically-based simulation model instead of real measurements. The simulation of three scenarios corresponding to constant, slow and fast input dynamics were compared. Average and maximum deviations were used as performance measures to evaluate and compare the models. On the basis of the comparisons, the topology of the best model was identified. The results show that moisture distribution in the drying bed could be well modelled using a neural network.